CrystEngComm

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

.

PAPER View Article OnlineView Journal

This journal is © The Royal Society of Chemistry 2015

aDepartment of Physics and Photon Science, Gwangju Institute of Science and

Technology, Gwangju, 500-712, Republic of KoreabDepartment of Electrical Engineering and Computer Science, Nagoya University,

464-8603, Japanc School of Materials Science and Engineering, Gwangju Institute of Science and

Technology, Gwangju 500-712, Republic of Koread Advanced Photonics Research Institute, Gwangju Institute of Science and

Technology, Gwangju 500-712, Republic of Koreae School of Information and Communications, Gwangju Institute of Science and

Technology, Gwangju 500-712, Republic of Korea

† Contributed equally to this work.

Cite this: DOI: 10.1039/c5ce00543d

Received 17th March 2015,Accepted 22nd June 2015

DOI: 10.1039/c5ce00543d

www.rsc.org/crystengcomm

Evolutionary growth of microscale singlecrystalline GaN on an amorphous layer by thecombination of MBE and MOCVD

Jung-Wook Min,†a Si-Young Bae,†b Won-Mo Kang,c Kwang Wook Park,d

Eun-Kyu Kang,e Bong-Joong Kim,c Dong-Seon Lee*e and Yong-Tak Lee*de

To integrate multiple functional devices on a chip, advances in epitaxial growth on heterosubstances are

required. We developed a method of combined epitaxial growth using molecular beam epitaxy (MBE) and

metal–organic chemical vapor deposition (MOCVD) as one approach to achieve an epitaxial layer on an

amorphous substrate. This two-stage combined growth can be used to grow binary gallium nitride (GaN)

on any thermally durable substance. The first MBE growth step provided effective nucleation with uniform

morphology. Meanwhile, the second MOCVD growth step resulted in improved crystalline quality. Detailed

analysis of grain-to-grain and layer-to-layer interfaces was performed with high-resolution transmission

electron microscopy (TEM) characterization. This study gives a deep understanding of the growth behavior,

thereby supporting the generation of perfect single crystalline GaN, enabling the realization of optoelec-

tronic GaN-based devices on an amorphous layer.

Introduction

GaN-based heteroepitaxial growth techniques have been suc-cessfully established in solid-state optoelectronics since thefirst demonstration of high brightness light-emitting diodes(LEDs).1,2 Nowadays, high quality GaN can be grown on Sisubstrates with promotion of large-scale GaN epitaxy on sub-strates over 8 inches in size, resulting in high yields of chipswith low cost.3 However, the strong light absorption by the Sisubstrate over the entire visible wavelength range is one ofthe critical drawbacks to its application in visible light-emitting devices.4 In efforts to address the unfavorable prop-erties of Si substrates, several research groups have reportedvertical LEDs with a Si substrate removal process. To over-come the limitations of existing substrates and explore novelapplications, various alternative substrates have been investi-gated, such as reflective metallic substrates, e.g. Ag, Cu, W,Ta, Mo, Nb, etc. or transparent amorphous substrates, e.g.

diamond, quartz, fused silica and so on.5–8 So far, few investi-gations have reported the direct growth of GaN on amorphoussubstrates via metal–organic chemical vapor deposition(MOCVD) since directly grown GaN typically possess weaklypreferred orientations and rough morphologies due to theabsence of appropriate assistant buffer materials.7,9 Themajority of efforts to grow GaN on amorphous substrateshave been attempted using electron cyclotron resonancemolecular beam epitaxy (ECR-MBE).6,8 The ECR cell providesremarkably enhanced N2 reactivity in forming a GaN layer onamorphous surfaces, and the ECR cell-assisted nitridationprocess in the initial stage improves both crystal quality andsurface morphology.10,11 However, ECR-MBE grown GaNexhibits poor crystal quality due to excessively high plasmapower.12

On the one hand, there is another alternative, called radiofrequency plasma-assisted MBE (PA-MBE), which allowsatomic-scale molecular beam epitaxial growth at low growthtemperature and follows high quality epitaxy. It also gener-ates sub-microscale polycrystalline grains of highly c-planeoriented GaN on amorphous substrates.11,13 Due to theVolmer–Weber (VW) dominant growth mode under the highV/III flux ratio of N-rich conditions, coalesced GaN columnscan be formed as pre-orienting layers with enhanced growthuniformity.6,8,11 Another advantage of PA-MBE is the easypolarity-based control of surface orientation, as either +c-plane orientation (Ga-polarity) or −c-plane orientation (N-polarity). In these processes, a conventional Ga-polar GaNlayer is introduced using a high-temperature AlN buffer layer

CrystEngComm

CrystEngCommPaper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

and a nitridation process on an Al-containing substrate,while an N-polar GaN layer is predominantly induced withoutan Al platform.14 In particular, the availability of N-polar GaNis a noteworthy feature since it provides reverse spontaneouspolarization along the direction opposite to Ga-polar GaN,thereby inducing a significant change in the energy band pro-files of a heterojunction.15 Furthermore, tailoring the bandprofile using an N-polar GaN layer has application in severalkinds of electronic/optoelectronic devices, e.g., field-effecttransistors (FET) using altered two-dimensional gases at theGaN/AlGaN interface, InGaN-based solar cells with reducedpiezoelectric polarization, and highly efficient LEDs withhigh indium incorporation as well as reduced efficiencydrop.16–18

On the other hand, there have been still several attemptsto grow GaN on substitute substrates using MOCVD.Recently, Choi et al. have reported an LED structure grownon an amorphous substrate.19 To improve the preferred ori-entation and crystalline quality, Choi et al. applied a Ti seedlayer as a pre-orienting layer. After formation of the preferredorientation, low temperature GaN growth and high tempera-ture local GaN growth were performed sequentially using aSiO2 mask layer on micro-sized openings via MOCVD.19 Thissuggested an easy method to tune the aspect ratio of locallygrown GaN simply by changing the growth conditions.20 Inaddition, there is still room, by optimizing the SiO2 maskopening size, to achieve the maximum crystalline quality ofthe local epitaxy. In the MOCVD system, it is difficult toobtain an optimized mask opening size due to the grainroughness beneath the SiO2 mask layer. The grain roughnessof MOCVD grown GaN is on the microscale, which is compa-rable to the size of the mask openings.7 Also, it is not easy toavoid these obstacles due to the nature of MOCVD growth.9

However, the MBE growth technique can induce a high den-sity of grains within the polycrystalline layer, whilst the grainsize becomes limited to the nanometer-scale with fairly uni-form size distribution.13 This small size of grains can beadvantageous for optimizing the mask opening size. At pres-ent, however, little attention has been given to structuralanalysis of the grains.

In this work, we investigated the distribution of grain sizeand the properties of the materials at each interface beforedetermination of the appropriate scale of local epitaxy. As anintermediate stage, this study tried to provide a deep under-standing of the GaN growth behavior on an amorphous layer.Furthermore, in order to easily achieve high quality GaNgrowth on a substitute substrate, we applied a combinedgrowth technique: PA-MBE growth (1st GaN) followed sequen-tially by MOCVD growth (2nd GaN). In the combined growthtechnique, an amorphous layer was applied as a substituteintermediate layer when growing the PA-MBE grown polycrys-talline GaN layer, to obtain a highly ±c-oriented GaN layer.We adopted Si3N4 grown on a quartz plate as the amorphouslayer. The thermal expansion coefficient of Si3N4 is betweenthose of GaN and quartz, so detrimental stress to the GaNlayer can be reduced during the cooling process. Additionally,

CrystEngComm

it is noteworthy that the Si3N4 layer can be easily depositedon any kind of thermally durable substrate.

Experimental1st GaN growth (MBE) procedure

Prior to epitaxial layer growth, a 500 nm thick Si3N4 layer wasdeposited on a quartz substrate using plasma enhancedchemical vapor deposition (PECVD). As precursors for reac-tion, NH3 and SiH4 gases were introduced into the PECVDchamber and subsequently, a Si3N4 layer was deposited at300 °C. Next, pieces of Si3N4/quartz samples (~5 cm2) weremounted on a molybdenum block using the indium pastemethod, and the block was loaded into the preparationchamber of modified PA-MBE (VG-V80). After pre-outgassingof the sample mounted on the molybdenum block at 300 °Cfor 30 min in the preparation chamber, the sample was trans-ferred to the growth chamber and thermally cleaned at 900°C for 30 min to remove residual native oxides on the samplesurface. Next, the nitridation process and the subsequent 1stGaN layer growth were performed on the Si3N4 layer in a PA-MBE growth chamber equipped with an RF plasma source(Arios IRFS 504) and in situ reflection high-energy electrondiffraction (RHEED) system at 20 kV. The 1st GaN layer wasgrown for 1 hour at 700 °C with plasma atmosphere gener-ated under 350 W of RF power and 1.0 sccm of N2. Byramping the growth temperature up to 800 °C, 3 × 10−7 mbarof Ga beam equivalent pressure (BEP) was introduced intothe sample without any buffer layer. With the growth process,a 700 nm thick polycrystalline GaN layer was grown, and thechamber pressure was kept at around 3.5 × 10−5 mbar. Dur-ing MBE growth, a RHEED (Arios 201) analysis technique wasapplied to perform in situ monitoring of the quality and ori-entation of the GaN layer.

2nd GaN growth (MOCVD) procedure

After completing the 1st GaN layer growth using MBE, thesample was transferred into the MOCVD reactor (MO260S,Sysnex, Korea) for the 2nd GaN layer growth. In this step, the2nd GaN layer was grown not only on the bare surface (pla-nar 2nd GaN) but also on the patterned surface of the 1stGaN layer using selective area growth (SAG-2nd GaN). Toform the patterned surface on the 1st GaN surface, a stripe-patterned (window/mask length = 3 μm/3 μm) SiO2 mask wasformed; first, a 200 nm thick SiO2 layer was deposited on the1st GaN layer by PECVD and then, the stripe openings forSAG were defined by photolithography and etched by reactiveion etching (RIE) using CF4/O2 gases. The samples wereheated to 1050 °C at 13.3 kPa under ambient hydrogen (H2).As soon as the growth temperature reached 450 °C, ammonia(NH3) was injected to reduce the thermal dissociation of the1st GaN layer. For the synthesis of the 2nd GaN layer, theflow rates of trimethylgallium (TMGa) and NH3 weremaintained at 100 sccm (165.5 μmol min−1) and 8 slm (357.1mmol min−1), respectively. The growth duration of the 2ndGaN layer was kept at 45 min.

This journal is © The Royal Society of Chemistry 2015

Fig. 1 Schematic of the growth procedure for (i) Si3N4 deposition on quartz wafer, (ii) 1st GaN layer growth via MBE, (iii) SiO2 stripe maskformation via PECVD, and (iv) 2nd GaN layer growth via MOCVD. Planar growth was performed on a bare 1st GaN template without maskformation (up) and SAG was performed on a stripe-patterned surface of the 1st GaN layer (down).

CrystEngComm Paper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

Fig. 1 illustrates the combined epitaxial growth procedurewith MBE and MOCVD. To eliminate any unpredictable fac-tors during the 2nd GaN layer growth step, both the planar2nd GaN layer growth and the 2nd GaN layer SAG wereperformed at the same time in the MOCVD reactor to studythe growth features under the same conditions.

Fig. 2 SEM images of PA-MBE grown 1st GaN surfaces on the amor-phous Si3N4 layer at growth times of (a) 5 min, (b) 10 min, (c) 40 min,and (d) 60 min. The inset of Fig. 2(d) is the in situ RHEED pattern duringMBE growth, where the polycrystalline orientation of the 1st GaN layerwas reflected. As growth time elapses, the size of nanoclustersincreased and the surface was almost fully covered by thesenanoclusters.

Sample treatment and characterization

To investigate the polarity of the 2nd GaN layer, 3 M KOHetching solution was prepared. The SAG-2nd GaN sample wasdipped into the KOH solution at 80 °C for 5 min. The surfacemorphology and atomic intensity were investigated by field-emission scanning electron microscopy with energy disper-sive X-ray (EDX) analysis (FE-SEM, Hitachi S-4700 EMAX). Toinvestigate the structural properties of the grown GaN layersas well as the amorphous layer, transmission electron micros-copy (TEM, Tecnai G2 F30 S-Twin) was carried out to obtaindark-field (DF) images, high-resolution (HR) images, andselected area diffraction (SAD) patterns at the interfaces. Forthe TEM measurement, samples were sliced by a dual-beamfocused ion beam (DB-FIB, NOVA 200). The crystal orienta-tions and crystalline quality of the grown samples were evalu-ated by X-ray diffraction (XRD, PANalytical X'Pert Pro). Fur-ther optical properties were characterized byphotoluminescence (PL) and cathodoluminescence (CL,MONO CL3+) which were used to perform spatial emissionmapping and spot-mode spectra measurement.

This journal is © The Royal Society of Chemistry 2015

Results and discussionI. Growth evolution of the 1st GaN layer via PA-MBE

In order to observe the initial growth behavior on the amor-phous Si3N4 layer during PA-MBE, we grew the samples bychanging the growth time. Fig. 2 presents the SEM images of

CrystEngComm

Fig. 3 (a) EDX characterization and (b) room temperature PLmeasurement of 1st GaN at increasing growth times of 5 min, 10 min,40 min, and 60 min. Initial nucleation was mainly attributed to theGaOx alloy nanoclusters and the main peak (E1) was blue-shifted by0.34 eV as the growth time increased up to 60 min.

CrystEngCommPaper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

the sample surfaces for growth times from 5 min to 60 min.We could observe that droplet-like nanoclusters were formedon the surface at the beginning (Fig. 2(a)). As the growth timeelapsed, the size of the nanoclusters increased and the sur-face was almost fully covered by these nanostructures (Fig.2(b) and (c)). Finally, it had closely packed hundreds-of-nano-meter-sized columnar grains with a randomly twisted in-plane direction (Fig. 2(d)). At the same time, we monitoredthe RHEED patterns during MBE growth via an in situRHEED gun mounted on the chamber, as shown in the insetof Fig. 2(d). Whilst no distinguished patterns were observedduring the nitridation and early growth processes (notshown), disconnected ring-like patterns appeared in the 1stGaN grown for 60 min. Bright spots in the RHEED patterncorrespond to (002), (004), (101), and (103) orientations,thereby indicating its polycrystalline nature.21

Note that MOCVD epitaxy on the lattice mismatched sub-strate (and/or amorphous layer) typically results in randomdistribution of nucleated three dimensional (3D) bulky GaN,i.e., it is quite difficult to grow a GaN film which covers thewhole amorphous surface.22 To understand the mechanismbehind the full surface coverage achieved with the MBEgrown GaN layer, we first performed EDX characterization foreach sample (Fig. 3(a)). N atoms were not detected at thebeginning of growth, but only Ga and O atoms were observed(5 min and 10 min). It means that the initial nucleation wasmainly attributed to the GaOx alloy nanoclusters. As thegrowth time elapsed, N atoms started to be detected, whilstO atoms slowly vanished (40 min and 60 min). We believethat these O atoms originated from native oxides on the sam-ple surface. The monotonic decrease of Si atoms was attrib-uted to intensity weakening of the EDX signal from the upperthicker structure. Hence, the GaOx nanoclusters might play akey role as a nucleation region to grow the upper columnarGaN structure. To confirm the role of GaOx nanostructures,we further implemented PL measurement on these sampleswith a He–Cd laser as shown in Fig. 3(b). One main peak (E1)was found at 2.02 eV for the 5 min sample and it was blue-shifted by 0.34 eV as the growth time increases up to 60 min,whereas the other satellite peak was located at 2.36 eV for the10 min and 40 min samples. It is speculated that these E1and E2 peaks originated from the stoichiometric variation ofIJGa2O3)n clusters (n = 1–10) as theoretically reported else-where.23 In addition, the remarkable blue-shifting behaviorof E1 quite well matches with the size effect of the Ga2O3

nanostructures.24

Fig. 4 SEM images of (a) planar 2nd GaN grown for 15 min, (b) itszoomed image of microscale platelet-shaped grains, (c) planar 2ndGaN grown for 45 min, and (d) its zoomed image of enlarged planargrains.

II. Growth morphology of the 2nd GaN layer via MOCVD

As for the next step of the 1st GaN growth, the 2nd GaNgrowth was implemented through MOCVD. Fig. 4 displaysthe SEM images of planar 2nd GaN grown for 15 min(Fig. 4(a) and (b)) and 45 min (Fig. 4(c) and (d)). It showsmicroscale platelet-shaped grains larger than the ones foundin the 1st GaN layer. The size of these platelet-shaped grainsincreased as the angle difference between one grain and an

CrystEngComm This journal is © The Royal Society of Chemistry 2015

CrystEngComm Paper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

adjacent grain becomes smaller. As can be seen in Fig. 4(b),for example, G1 and G2 were composed of rather enlargedplanar grains with a small angle difference (Δθ1), whereasabrupt distortion was formed at the interface between G2and G3 with a large angle difference (Δθ2). The increasedgrowth time induces much more chances of the grains com-ing into contact with the other grains with a larger angle dif-ference; therefore, planar grains get reduced and 3D micro-structures become dominant, as shown in Fig. 4(c).Nevertheless, some particular planar grains could be grownwith a grain size of ~3.2 μm, as shown in Fig. 4(d). Hence,the coherence of the grain-to-grain edge angle is quite impor-tant to the evolution of the two dimensional (2D) layer.

To reduce the chances of bringing the microscale grainsinto contact with neighboring grains, we performed SAG witha stripe-shaped mask, thereby reducing the degree of free-dom on the in-plane growth direction by half (Fig. 5). As aresult, a controlled morphology along the stripe window wasachieved and the size of the platelet-shaped grains becameapproximately two times larger than that of the planar 2ndGaN. Furthermore, some cracks observed on the 1st GaNlayer were covered by coalescence above crack boundaries, asshown in Fig. 5(a). It indicates that microstructural disloca-tions could be reduced by strong lateral growth. The averagegrowth rate of the planar 2nd GaN layer was ~5 μm h−1 whichis larger than twice of the one obtained with typical planarGaN on sapphire (~2.2 μm h−1) under the same growth condi-tions, indicating that vertical growth was enhanced by island-like Stranski–Krastanov (SK) mode growth. In contrast, theSAG-2nd GaN layer had a slightly reduced growth rate of ~4.4μm h−1, indicating enhancement of lateral growth. However,the striped-SAG also resulted in the formation of distorted3D microstructures as the growth time increased due to thecompetition of individual grains which still have a twistedpreferred orientation along the in-plane direction (Fig. 5(b)).

III. Microstructural TEM analysis

In order to study the structural properties of the grownlayers, TEM analysis was carried out. Fig. 6(a)–(c) show theDF TEM images taken from the 1st GaN layer, the planar 2ndGaN layer, and the SAG-2nd GaN layer, respectively. InFig. 6(a), the cross-sectional view shows that the 1st GaNlayer is composed of a few hundreds-of-nanometer columnar

This journal is © The Royal Society of Chemistry 2015

Fig. 5 SEM images of SAG-2nd GaN grown by MOCVD for (a) 15 minand (b) 45 min. Controlled morphology along the stripe window wasachieved and grain size became two times larger than that of planar2nd GaN (Fig. 4(b) and (d)).

grains on the amorphous Si3N4 layer. Typically, it showsheights of ~700 nm and diameters of ~250 nm. On the otherhand, the MOCVD growth apparently induced much enlargedgrains with sizes of a few micrometers on the planar 2ndGaN and SAG-2nd GaN layers (Fig. 6(b) and (c)). The grainenlargement is believed to be mainly caused by the long sur-face diffusion length of adatoms and the altered growthmode via MOCVD. In particular, the growth temperature ofMOCVD which is higher by 250 °C compared to that of MBEaffected the Si3N4 layer, thereby resulting in deformation andre-crystallization of the amorphous Si3N4 layer (Fig. 6(b) and(c)), while the Si3N4 layer did not suffer from any deforma-tion during the 1st GaN growth step (Fig. 6(a)). Note thatGaN layers were often grown on a few nanometers of Si3N4

layer to reduce the TDs for GaN/Si.25 In other words, this caneasily explain why TD reduction was possible at theGaN/Si3N4 interface, providing additional GaN growth siteson the Si3N4 intermediate layer. We speculate that the higherby 250 °C growth temperature of MOCVD enables the inter-diffusing behavior of the regrown GaN layer. Additionally,during the cooling process, thermal shrinkage critically dam-aged the crystallized Si3N4 layer, thereby forming severecracks. Note that the CTEs of GaN, Si3N4, and quartz are 5.6,3.3, and 0.3 × 10−6 K−1, respectively. In detail, during theMBE growth, the 1st GaN layer was epitaxially formed on theSi3N4 layer at a growth temperature of 800 °C. At this time,most of the deformations were generated at the outermostpart of the 1st GaN layer, which resulted in the formation ofsevere cracks of the 1st GaN layer. Next, during the MOCVDgrowth, the growth temperature was higher by 250 °C com-pared to the MBE growth temperature, and as a result, Gainter-diffusion occurred from the 1st GaN layer to the Si3N4

layer. Then, the crystallized Si3N4 layer suffered from strongtensile strain due to the large thermal mismatch with thequartz substrate during the cooling step. Considering theMOCVD cases, voids were found, which is possibly due to thefaster growth of neighboring grains as well as cracks duringcooling. It is observed that the planar 2nd GaN layer hadvivid threading dislocations (TD) from the top of the 1st GaNto the 2nd GaN layer as well as bunches of defects, which areshown in Fig. 6(b). Meanwhile, the SAG-2nd GaN layershowed rather reduced TDs, as seen in Fig. 6(c). Despite thereduction of TDs in the SAG-2nd GaN layer, elongated lateraldefects were formed on planes parallel to the substrates, asshown in the selected red-dashed rectangle in Fig. 6(c). In avast number of cases, these are accepted as basal-plane stack-ing faults (BSFs). Fig. 6(d)–(f) display the SAD patterns of the1st GaN layer, planar 2nd GaN layer, and SAG-2nd GaN layer,respectively. Most of the polycrystalline properties wereproven by the ring-like patterns of the 1st GaN layer (Fig. 6(d)).Quasi-discrete spots on the ring-like pattern are caused by thetwisted in-plane orientation of the neighboring grains. Fig. 6(e)shows the SAD pattern of the planar 2nd GaN layer in themiddle of a grain. Single crystalline wurtzite GaN was con-firmed by the clear six-fold symmetry. This indicates thatthe enlarged GaN grown by MOCVD effectively enhances the

CrystEngComm

Fig. 6 Cross-sectional dark-field TEM images of (a) 1st GaN, (b) planar 2nd GaN, and (c) SAG-2nd GaN. Their corresponding SAD patterns areshown in Fig. 6(d)–(f), respectively. The high density of defects including TDs in planar 2nd GaN was reduced by SAG. Each microscale grain pos-sess single GaN crystallinity within the middle of the grains.

CrystEngCommPaper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

crystalline quality, but the single crystalline properties areonly effective within a single micrometer-sized grain. TheSAG-2nd GaN layer also showed single crystallinity in themiddle of a grain (Fig. 6(f)). Hence, we believe that muchsmaller scale local epitaxy such as nano-scale mask patternsmight open more opportunities to improve the quality ofcrystal arrays.

To understand the growth evolution of GaN on an amor-phous layer, it is important to investigate the structural prop-erties of the interfaces, e.g., grain-to-grain and layer-to-layer.Fig. 7(a) and (b) show the HR TEM images of the 1st GaNlayer near the Si3N4 layer and the middle of the grain-to-graininterface after MBE growth. Here, we could observe the directevidence of Ga2O3 nanoclusters on the Si3N4 layer, asdiscussed from EDX characterization in Fig. 3. The insets ofFig. 7(a) show the fast Fourier transformation (FFT) on localregions, displaying dispersed FFT patterns of the crystallinedisorder induced by defects and twisted orientation. Forexample, the underlying Si3N4 layer has a quite blurred FFTpattern, indicating amorphous properties without any re-crystallization during MBE growth. Meanwhile, differentatomic arrangements on the left and right sides of the grainboundary (GB) were found as wurtzite (WZ) and partiallyzincblende (ZB-like) structures. Much clear evidence of thesemixed crystal features was observed around the GB in the

CrystEngComm

middle of the 1st GaN layer in Fig. 7(b), where WZ and ZB-like atom stacking is seen in the red boxes. After MOCVD re-growth, on the other hand, the Si3N4 region became re-crys-tallized, as shown in the hexagonal arrangement of the HRimage as well as FFT patterns (bottom) in Fig. 7(c). It mightbe attributed to deep inter-diffusion, indicating that the Gaatoms from the 1st GaN layer diffused into the Si3N4 layer.26

We still could find Ga2O3 regions with a mixed structure, asshown in the upper FFT pattern in Fig. 7(c). Fig. 7(d) showsthe HRTEM image taken from the interface between 1st GaNand 2nd GaN layers. Although the spots of the planar 2ndGaN layer were clearer than those of the 1st GaN layer, a highdensity of stacking-fault-like defects was also determinedthrough the dispersed FFT pattern (inset of Fig. 7(d)).

IV. Optical properties and residual strain

To examine the spatial arrangement of crystallinity, CL measure-ment was carried out. Fig. 8(a) shows a cross-sectional SEMimage of the planar 2nd GaN layer. Its spatial CL images werealso shown in panchromatic (Fig. 8(b)) and monochromaticviews (Fig. 8(c)). No luminescence was observed at the 1st GaNlayer due to low crystallinity caused by the high density ofdefects. In contrast, randomly formed local areas on the 2ndGaN layer showed near-band-edge emission. As proven by the

This journal is © The Royal Society of Chemistry 2015

Fig. 7 High-resolution TEM images and corresponding FFT patterns(insets) of (a) 1st GaN layer near a Si3N4 layer, (b) the middle of thegrain-to-grain interface, (c) 1st GaN layer near a re-crystallized Si3N4

layer, and (d) the interface between 1st GaN and planar 2nd GaN. Poly-crystalline properties were observed at grain-to-grain interfaces for allsamples. The FFT pattern became vivid after conducting MOCVDgrowth, indicating improved crystallinity. Red-lined rectangles showthe defects in grains and red circles show Ga2O3 regions with mixedstructures.

CrystEngComm Paper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

TEM images in Fig. 6, GB and TD can play important roles asnon-radiative recombination centers at the interfaces. The weakemission in the monochromatic CL image at 3.42 eV indicatesthat the planar 2nd GaN layer has poor crystal quality. Fig. 8(d)

This journal is © The Royal Society of Chemistry 2015

Fig. 8 (a) Cross-sectional SEM image of planar 2nd GaN. Its correspondSpot-mode CL spectra measured on P1–P4 in Fig. 8(a). (e) Cross-sectional S(g) monochromatic CL emission images. (h) Spot-mode CL spectra measure

shows the spot-mode CL spectra of each position from P1 to P4,as shown in Fig. 8(a). No distinguishable emission was mea-sured around the 1st GaN layer (P1) and 2nd GaN layer (P2). Incontrast, near-band-edge emission (3.44 eV) was observed in themiddle of the microscale-sized grain (P3 and P4). The domi-nance of the shorter emission energy at 3.37 eV and the emis-sion broadening are possible due to the high density of BSFmixtures such as I1-, I2-, and E-BSF.27 On the other hand,Fig. 8(e) shows a cross-sectional SEM image of the SAG-2ndGaN layer. Its panchromatic (Fig. 8(f)) and monochromatic(Fig. 8(g)) CL images also appeared in the same manner as thatof the planar 2nd GaN layer. Emissive regions were clearlyexpanded in comparison to those of the planar 2nd GaN layer.Spread growth from the window opening to both outer sideswas observed in a bending manner. This bending behaviorclearly reduces TDs, as reported elsewhere, thereby enhancingband-edge emission distribution in Fig. 8(g).28 Fig. 8(h) presentsthe spot-mode CL spectra of each position from S1 to S4, asshown in Fig. 8(e). No emissive regions (S1) existed at the GBon the window opening. Above the masking area, a band-edgeemission at 3.43 eV was dominantly identified, which wasslightly red-shifted due to weak tensile strain caused by SAG. Itwas also found that there is a high density of BSF mixtures asevidenced by the peak broadening at around 3.35 eV, where I2BSF is prominent. However, the intensity of the band-edge emis-sion was higher than that for the BSF centers as well as the pla-nar 2nd GaN layer, which indicates that SAG was a quite effec-tive growth method for improving the crystal quality. Therefore,natural growth competition between neighboring grains can becontrolled even on GaN on an amorphous substrate.

XRD measurement could provide further reliable examina-tion of the crystalline properties. Fig. 9(a) shows the 2θ-scansof the grown samples, where dominant peaks were detected

CrystEngComm

ing (b) panchromatic and (c) monochromatic CL emission images. (d)EM image of selective 2nd GaN. Its corresponding (f) panchromatic andd on S1–S4 in Fig. 8(e).

CrystEngCommPaper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

at 34.8°, corresponding to (0002) orientation. Two other satel-lite peaks located at around 32° and 45° are representative ofthe 1st GaN layer, indicating (101̄0) orientation and reflection(asterisk) from the aluminium sample holder. This indicatesthat the 1st GaN layer contains a few mixed orientations, asdiscussed in Fig. 7. However, when we grew the 2nd GaNlayer by MOCVD, the satellite peaks found in the 2θ-scansalmost vanished, and instead, the signal intensity of the peakcorresponding to (0004) orientation was highly enhanced andthe peak was observed as a family of (0002) peaks. In addi-tion, the full width at half maximum (FWHM) of the regrown2nd GaN layer became narrower than that of the 1st GaNlayer. Hence, the combined epitaxy with MOCVD and MBEcertainly contributes to the crystal ordering, shown as theselection of the preferred orientation along the c-axis direc-tion, as well as improved crystal quality.

The effect of strain on epitaxial growth is also of importancesince it could be related to wafer bowing, e.g., a large CTEmismatch results in the formation of severe cracks due to hugetensile strain, as can be seen in typical GaN/Si. The Raman spec-tra exhibit the remaining strain of the top layer for the 1st GaN

CrystEngComm

Fig. 9 (a) XRD 2θ-scans and (b) Raman spectra of grown layers. Theslightly mixed orientation of 1st GaN became a (0002) orientation bygrowing via the MOCVD growth step. Strain was fully relaxed on 1stGaN and planar GaN, while SAG-2nd GaN suffered from slight tensilestrain.

and 2nd GaN layers, as shown in Fig. 9(b). Note that the E2mode is related to a displacement in the c-plane, and the A1 lon-gitudinal optical IJLO) mode comes from atomic relative motionsalong the c-axis direction.29 The E2 peak was observed at 567.2cm−1 in both the 1st GaN and planar 2nd GaN layers. Since thisindicates a nearly strain-free state (567 ± 0.1 cm−1), residualstresses in the 1st GaN and planar 2nd GaN layers were negligi-ble. For the SAG-2nd GaN layer, the E2 peak was located at566.6 cm−1, indicating a slight tensile strain of 0.161 GPaobtained using the equation σ = Δω/4.3 cm−1.30 Hence, the lat-eral growth enhancement by SAG affected the degree of strainin the GaN layer, in a manner quite similar to that of otherreports elsewhere.28 The emergence of E1 at 532 cm−1 is pre-sumably attributed to the stripe morphology of the SAG-2ndGaN layer since the E1 mode of the Raman spectra can beover-susceptible based on the surface geometry.30,31 TheA1IJLO) peaks also reflect the strain status in a similar manner.The A1IJLO) peak of the planar 2nd GaN layer is located at 736cm−1 without dispersion properties, and this well matches withthat of the strain-free GaN layer (736.5 ± 0.2 cm−1), while theA1IJLO) peak of the selectively grown GaN layer is at 734.1cm−1, indicating slight tensile strain due to a short Raman fre-quency shift.30 However, broad dispersion is shown in the 1stGaN layer. It is speculated that the peak broadening is attrib-uted to unintentional inter-diffusion of Si atoms from Si3N4 tothe 1st GaN layer at the interface. Due to the coupling of theLO phonon with the overdamped plasmon, the A1IJLO) peakwas broadened and almost vanished with the increase in Siconcentration of the 1st GaN by inter-diffusion.32

V. Polarity study by chemical etching

Although both the planar and SAG-2nd GaN layers exhibited adominant c-plane orientation from XRD measurement, stud-ies about this in more detail are required since the grownstructures are composed of distorted microstructural grains.Hence, the grown samples were dipped into 3 M KOH solutionat 80 °C. Fig. 10(a) and (b) display the surface morphology ofthe 1st GaN layer after KOH etching. The procedure revealedthat grain boundaries tend to be etched and hexagonal grainswere almost retained. Note that N-polar GaN typically suffersfrom a strong chemical reaction compared to Ga-polar GaN,and therefore is rapidly etched.33 It indicates that individualgrains usually have Ga polarity, whilst grain boundaries haveN-exposed polarity. This quite well matches with the TEMobservation from Fig. 7(a) and (b), where ZB-like structureswere frequently found at the grain-to-grain interfaces. Further-more, the top surfaces of each grain were partially etched,indicating that the middle of the grain has slightly mixedpolar properties despite dominance of Ga-polarity.

Fig. 10(c) and (d) show the surface morphology of the SAG-2nd GaN layer after KOH etching. The morphology of thegrains changed to a pyramid-like shape composed of severalneedle-like structures corresponding to inversion domains(IDs) (Fig. 10(d)).34 It was also revealed that N-exposed regionswere strongly etched. We observed that several microscale

This journal is © The Royal Society of Chemistry 2015

Fig. 10 SEM images of (a) 1st GaN grown by MBE, (b) its zoomedimage, (c) SAG-2nd GaN, and (d) its zoomed image after chemicaletching in KOH solution. 1st GaN hexagonal grains are almost retained,but grain boundaries tend to be etched. SAG-2nd GaN shows a well-defined pyramid-like shape with columnar defects of IDs.

CrystEngComm Paper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

grains were stacked, where the grain interfaces might haveeasily mixed polarity as well as non-radiative components.Hence, these distorted grains might be easily etched during KOHetching. Some needle-like IDs were also evidence of local Gapolarity. Generally, IDs come from the interface to the layer sur-face with narrow and straight columnar structures. This is causedby the growth rate difference between Ga- and N-polar GaN. Hav-ing a higher growth rate, Ga-polar GaN propagates to the entiresurface from the interface and emerges at the centers of areaswith opposite N-polarity. After dipping into KOH solution, it isexpected that Ga-polarity IDs are etched much slower than thesurrounding areas of N-polarity. Finally, Ga-polarity IDs have astraight and narrow columnar shape, whilst the rest of the areaschanged to a pyramidal shape.34–36

Note that the conventional Ga-polar GaN layer is grown onthe Al covered surface of an AlN buffer layer or a sapphireIJAl2O3) substrate. At the interface, the higher bonding energyof Al–N (compared with Ga–N) determines the Ga polarity ofGaN by its bonding configuration. In this case, N polarity iskinetically unfavorable on the Al-containing surface.14,37 Onthe other hand, N polarity can be induced without an Al plat-form.38 In our case, the 1st GaN layer could be clustered onGa2O3 depleted regions, i.e., around Ga2O3 nanoclusters onSi3N4, thereby inducing N polarity as shown in the results ofKOH etching for the 1st GaN layer. In addition, N-polar GaNcould be further grown during MOCVD growth since the adja-cent grains do not have ordered height, thereby inducingpolarity inversion as a result of coalescence.39 Therefore, wecould conclude that the 2nd GaN layer has mixed polaritywith Ga- and N-polar GaN, where N polarity was usuallyenhanced around the grain boundaries.

VI. Proposed growth model and further prospects

Based on the analysis of GaN on an amorphous layer so far,here we propose a growth model, as illustrated in Fig. 11.

This journal is © The Royal Society of Chemistry 2015

Although it is intentionally exaggerated for our purpose, webelieve that it might be quite helpful to understand theresults shown in this study as well as to improve the crystalquality. The procedures are simply described in Fig. 11 as fol-lows: (i) Ga2O3 clustering, (ii) GaN nucleation, (iii) local GaNcoalescence, (iv) columnar growth via MBE, and (v) laterallyenhanced growth via MOCVD. The origin of the O source atthe beginning of GaN growth is not clear at this time. Wespeculate that it might be attributed to the native oxides onthe Si3N4/quartz sample surface. Due to chemical reaction ofSi3N4 and quartz in HF solution, a conventional cleaning pro-cess using HF solution should be avoided in the sample prep-aration step, so that native oxides would not be fully removedduring the thermal cleaning process in the MBE chamber.Then, Ga2O3 nanoclusters are formed on the surface afterexposure to Ga. Finally, Ga-polar GaN is formed usually onGa2O3, while N-polar GaN mainly originates from direct GaNclustering on the amorphous surface. Nitridation was performedat a rather longer time (1 hour), higher plasma power (350 W),and also higher substrate temperature (700 °C) than those for atypical nitridation process and epitaxy growth on sapphire or aGaN template.10 As a result, strong N2 reactivity is provided,forming initial GaN clusters on the amorphous Si3N4 layer.21

These clusters on the amorphous interface become N-polar GaNseeds on non-Ga2O3 regions, i.e., around Ga2O3 nanoclusters.Then, local coalescence occurs among GaN nucleators sincetheir distances are quite close by <100 nm. Further, nano-columns were formed by MBE growth, where the grown struc-tures were governed by VW growth mode, thereby inducingcolumnar growth.40–42 Finally, one large grain was formed onthese several nanocolumns and further lateral growth wasenhanced by MOCVD. When the edges of individual grainsmaintain their preferred orientation, an enlarged planar grain isformed with a six-fold hexagonal geometry. On the other hand,when the edges do not have coherent angles, this results in theformation of distorted grains with grain boundaries wheremixed crystal structures are frequently found. In the case ofMOCVD growth, mixed crystal structures such as wurtzite andzincblende mixtures are reduced, but single crystalline featuresin the middle of the microstructures are also observed.

As evident from these findings, we could suggest twoapproaches to improve the crystal qualities of GaN on anamorphous layer. One is to deposit (or insert) a new layerwith a preferred orientation before growth to induce coherentgrain edges. By considering material parameters such as in-plane lattice and thermal stabilities, Ti, Hf, and graphenecould be good candidates as reported elsewhere.19,43–45 Thesepreferred orienting layers can not only induce single crystal-line grains, but also enable further growth of an enlarged 2Dlayer. The other approach is to grow nanowire structures byusing smaller hole-shaped arrays such as nanoscale SAGstructures. Recently, there have been several reports toimprove the crystal quality by nanoscale epitaxy and for therealization of actual devices.26,46 These nanoscale SAG can bequite of interest for GaN growth on amorphous SAG sincesingle crystalline microstructures could be generated, as

CrystEngComm

Fig. 11 Schematics of the growth model and procedures for (i) Ga2O3 clustering, (ii) GaN nucleation, (iii) local GaN coalescence, (iv) columnargrowth via MBE, and (v) laterally enhanced growth via MOCVD.

CrystEngCommPaper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

evident in this study. Furthermore, these array-type GaNnanowires can suffer from much less biaxial strain than 2Dfilms since they are grown only on small localized regions.

Conclusion

Two-stage combined epitaxy was carried out with MBE andMOCVD to grow GaN on an amorphous layer. To obtain anadoptable layer on any amorphous substrate, a Si3N4 layer isdeposited and a quartz substrate is used as a thermally dura-ble substance. The 1st GaN layer growth via MBE provided anepitaxial buffer layer composed of grains of hundreds ofnanometers with planar surface morphology. Despite thepolycrystalline properties of the 1st GaN layer, it has a highlypreferred orientation along the c-plane on the amorphouslayer. The 2nd GaN layer growth via MOCVD improved thecrystalline properties, producing grains of a few micrometers.A mixed polarity of grown GaN was determined by XRD mea-surement as well as KOH chemical treatment. The growthmode was also changed from VW mode to SK mode, therebyincreasing the growth rate and reducing the density of thegrain boundary. This approach resulted in a strong andextended band-edge emission of the SAG-2nd GaN layerwithin a micro-sized grain, indicating its single crystallinequality, while high defect densities such as stacking faultsand threading dislocations were found for the planar 2ndGaN layer. A high-resolution TEM study was also performedat grain-to-grain and layer-to-layer interfaces. The results ofthe characterization could give us a much deeper under-standing regarding the availability of single crystalline GaN

CrystEngComm

on novel amorphous substrates. Therefore, this study couldbe a guide to grow much improved single crystals on othersubstances for the realization of optoelectronic GaN-baseddevices in the near future.

Acknowledgements

This research was supported by the Basic Science ResearchProgram through the National Research Foundation of Korea(NRF) funded by the Ministry of Education (NRF-2013R1A1A2A10006632) and the Core Technology Develop-ment Program for Next-Generation Energy of the ResearchInstitute for Solar and Sustainable Energies (RISE), GIST.

Notes and references

1 M. R. Krames, O. B. Shchekin, R. Mueller-Mach, G. O.

Mueller, L. Zhou, G. Harbers and M. G. Craford, J. Disp.Technol., 2007, 3, 160–175.

2 S. Pimputkar, J. S. Speck, S. P. DenBaars and S. Nakamura,

Nat. Photonics, 2009, 3, 180–182.

3 D. Zhu, D. J. Wallis and C. J. Humphreys, Rep. Prog. Phys.,

2013, 76, 106501.

4 M. A. Green and M. J. Keevers, Prog. Photovoltaics, 1995, 3,

189–192.

5 S. Inoue, K. Okamoto, T. Nakano, J. Ohta and H. Fujioka,

Appl. Phys. Lett., 2009, 91, 201920.

6 K. Yamada, H. Asahi, H. Tampo, Y. Imanishi, K. Ohnishi

and K. Asami, Appl. Phys. Lett., 2001, 78, 2849–2851.

This journal is © The Royal Society of Chemistry 2015

CrystEngComm Paper

Publ

ishe

d on

22

June

201

5. D

ownl

oade

d by

Kw

ang

Ju I

nstit

ute

of S

cien

ce a

nd T

echn

olog

y on

15/

07/2

015

12:5

1:50

. View Article Online

7 P. Hageman, J. Schermer and P. Larsen, Thin Solid Films,

2003, 443, 9–13.

8 M. Hiroki, H. Asahi, H. Tampo, K. Asami and S. Gonda,

J. Cryst. Growth, 2000, 209, 387–391.

9 S.-E. Park, D.-J. Kim, S. Woo, S.-M. Lim and B. O, J. Cryst.

Growth, 2002, 242, 383–388.

10 T. Araki, H. Kagatsume, T. Noguchi, T. Maruyama and Y.

Nanishi, Phys. Status Solidi C, 2003, 0, 200–204.

11 S. Fan, S. Zhao, X. Liu and Z. Mi, J. Vac. Sci. Technol., B:

Nanotechnol. Microelectron.: Mater., Process., Meas., Phenom.,2014, 32, 02C114.

12 W. C. Hughes, W. H. Rowland Jr., M. A. L. Johnson, S.

Fujita, J. W. Cook Jr., J. F. Schetzina, J. Ren and J. A.Edmond, J. Vac. Sci. Technol., B, 1995, 13, 1571–1577.

13 G. Nouet, P. Ruterana, H. Tampo and H. Asahi, J. Phys.:

Condens. Matter, 2002, 14, 12697.

14 K. Xu, N. Yano, A. Jia, A. Yoshikawa and K. Takahashi, Phys.

Status Solidi B, 2001, 228, 523–527.

15 A. Konar, A. Verma, T. Fang, P. Zhao, R. Jana and D. Jena,

Semicond. Sci. Technol., 2012, 27, 024018.

16 M. H. Wong, Y. Pei, T. Palacios, L. Shen, A. Chakraborty,

L. S. McCarthy, S. Keller, S. P. DenBaars, J. S. Speck andU. K. Mishra, Appl. Phys. Lett., 2007, 91, 232103.

17 J.-Y. Chang and Y.-K. Kuo, J. Appl. Phys., 2012, 112, 033109.

18 S. Keller, N. A. Fichtenbaum, M. Furukawa, J. S. Speck, S. P.

DenBaars and U. K. Mishra, Appl. Phys. Lett., 2007, 90, 191908.19 J. H. Choi, A. Zoulkarneev, S. I. Kim, C. W. Baik, M. H. Yang,

S. S. Park, H. Suh, U. J. Kim, H. Bin Son, J. S. Lee, M. Kim,J. M. Kim and K. Kim, Nat. Photonics, 2011, 5, 763–769.

20 B. O. Jung, S.-Y. Bae, Y. Kato, M. Imura, D.-S. Lee, Y. Honda

and H. Amano, CrystEngComm, 2014, 16, 2273–2282.

21 M. Zhong, F. Qin, Y. Liu, C. Wang, J. Bian, E. Wang, H.

Wang and D. Zhang, J. Alloys Compd., 2014, 583, 39–42.

22 B. Leung, J. Song, Y. Zhang and J. Han, Adv. Mater.,

2013, 25, 1285–1289.

23 A. B. Rahane, M. D. Deshpande and S. Chakraborty, J. Phys.

Chem. A, 2012, 116, 10559–10565.

24 S. S. Kumar, E. J. Rubio, M. Noor-A-Alam, G. Martinez, S.

Manandhar, V. Shutthanandan, S. Thevuthasan and C. V.Ramana, J. Phys. Chem. C, 2013, 117, 4194–4200.

25 M. Haeberlen, D. Zhu, C. McAleese, M. J. Kappers and C. J.

Humphreys, J. Phys.: Conf. Ser., 2010, 209, 012017.

26 J. H. Choi, E. H. Cho, Y. S. Lee, M. Shim, H. Y. Ahn, C. Baik,

E. H. Lee, K. Kim, T. Kim and S. Kim, Adv. Opt. Mater.,2014, 2, 302–302.

This journal is © The Royal Society of Chemistry 2015

27 J. Lähnemann, U. Jahn, O. Brandt, T. Flissikowski, P. Dogan

and H. T. Grahn, ArXiv14051261 Cond-Mat, 2014.

28 Y. Kato, S. Kitamura, K. Hiramatsu and N. Sawaki, J. Cryst.

Growth, 1994, 144, 133–140.

29 M. S. Liu, L. A. Bursill, S. Prawer, K. Nugent, Y. Tong and G.

Zhang, Appl. Phys. Lett., 1999, 74, 3125–3127.

30 C. Kisielowski, J. Krüger, S. Ruvimov, T. Suski, J. Ager III, E.

Jones, Z. Liliental-Weber, M. Rubin, E. Weber and M.Bremser, Phys. Rev. B: Condens. Matter Mater. Phys.,1996, 54, 17745.

31 S. Nagarajan, O. Svensk, M. Ali, G. Naresh-Kumar, C. Trager-

Cowan, S. Suihkonen, M. Sopanen and H. Lipsanen, Appl.Phys. Lett., 2013, 103, 012102.

32 T. Kozawa, T. Kachi, H. Kano, Y. Taga, M. Hashimoto, N.

Koide and K. Manabe, J. Appl. Phys., 1994, 75, 1098–1101.

33 D. Zhuang and J. H. Edgar, Mater. Sci. Eng., R, 2005, 48,

1–46.

34 J. Jasinski, Z. Liliental-Weber, Q. S. Paduano and D. W.

Weyburne, Appl. Phys. Lett., 2003, 83, 2811–2813.

35 V. Potin, P. Ruterana and G. Nouet, Mater. Sci. Eng., B,

1999, 59, 173–176.

36 L. T. Romano, J. E. Northrup and M. A. O'keefe, Appl. Phys.

Lett., 1996, 69, 2394–2396.

37 A. Yoshikawa and K. Xu, Thin Solid Films, 2002, 412, 38–43.

38 M. Stutzmann, O. Ambacher, M. Eickhoff, U. Karrer, A. Lima

Pimenta, R. Neuberger, J. Schalwig, R. Dimitrov, P. J. Schuckand R. D. Grober, Phys. Status Solidi B, 2001, 228, 505–512.

39 J. Song, G. Yuan, K. Xiong, B. Leung and J. Han, Cryst.

Growth Des., 2014, 14, 2510–2515.

40 V. Consonni, M. Hanke, M. Knelangen, L. Geelhaar, A.

Trampert and H. Riechert, Phys. Rev. B: Condens. MatterMater. Phys., 2011, 83, 035310.

41 T. Stoica, E. Sutter, R. J. Meijers, R. K. Debnath, R. Calarco,

H. Lüth and D. Grützmacher, Small, 2008, 4, 751–754.

42 R. K. Debnath, R. Meijers, T. Richter, T. Stoica, R. Calarco

and H. Lüth, Appl. Phys. Lett., 2007, 90, 123117.

43 K. Okamoto, S. Inoue, T. Nakano, J. Ohta and H. Fujioka,

J. Cryst. Growth, 2009, 311, 1311–1315.

44 T. Araki, S. Uchimura, J. Sakaguchi, Y. Nanishi, T.

Fujishima, A. Hsu, K. K. Kim, T. Palacios, A. Pesquera and A.Centeno, others, Appl. Phys. Express, 2014, 7, 071001.

45 J. W. Shon, J. Ohta, K. Ueno, A. Kobayashi and H. Fujioka,

Sci. Rep., 2014, 4.

46 K. Kishino, K. Nagashima and K. Yamano, Appl. Phys.

Express, 2013, 6, 012101.

CrystEngComm