Solutions for heteroepitaxial growth of GaN and
their impact on devices

M A R K U S K A M P

Department of Optoelectronics, University of Ulm, 89069 Ulm, Germany

(E-mail: [e-mail])

Abstract. GaN technology relies on highly mismatched heteroepitaxial growth, mainly on sapphire or SiC

substrates, and therefore su€ers from 10

9

to 10

10

threading dislocations per cm

2

. The origin and the

deteriorating in¯uence of the extremely high dislocation densities are analyzed with regard to the speci®c

circumstances of GaN technology. Various attempts to cope with heteroepitaxial growth are discussed,

from the use of nucleation layers to the growth on GaN single bulk crystals. Special focus is put on the

impact of the approaches on the device performance.

Key words: dislocations, GaN, heteroepitaxy, laser, LEDs

1. Introduction

GaN based materials are today's fastest developing III±V compound semi-

conductor technology. The excellent optical and electrical properties, the

wide direct bandgap, the thermal, mechanical, and chemical robustness make

GaN based semiconductors the superior material system for optoelectronic

devices (LEDs, laser, photodetectors) in the UV to visible range. Addition-

ally, electronic devices such as GaN based ®eld e€ect transistors (FETs) and

heterobipolar transistors (HBTs) o€er new applications in high power, high

frequency microelectronics. Various opto- and micro-electronic devices are

either already established or approaching the markets. Despite the tremen-

dous success this technology still su€ers mostly from the lack of a perfect

substrate and therefore has to cope with strongly mismatched heteroepitaxial

growth. Di€erences in lattice constants as well as thermal expansion coe-

cients result in about 10

9

±10

10

dislocations/cm

2

thus limiting device perfor-

mance and lifetime.

This paper provides an introduction to general issues of heteroepitaxial

growth and the generation of dislocations, both with special regard to GaN

technology and the impact on device performance. Potential substrates as

well as various techniques for the reduction of dislocation densities are

elaborated. Di€erent approaches to improve the layer properties, from low

temperature nucleation layers to homoepitaxial growth, are discussed.

In particular the new results on homoepitaxial growth on GaN single bulk

crystals provide new standards in GaN material quality. The exceptional

Optical and Quantum Electronics 32: 227±248, 2000.

Ó

2000 Kluwer Academic Publishers. Printed in the Netherlands.

227

quality is determined by a reduction of the photoluminescence linewidth

from 5 to 0.1 meV and a reduced XRD rocking curve width from 400 to

20 arcsec. The outstanding material quality provided new insights into fun-

damental material parameters (e.g. lattice parameters, excitonic binding en-

ergies, etc.) being not accessible by heteroepitaxial growth.

2. Potential substrates for GaN technology

Despite the fast and outstanding achievements of GaN based optical and

electrical devices, the technology still su€ers from strongly mismatched he-

teroepitaxial growth. GaN substrates pulled from a melt are not available,

yet. Predicted temperatures and pressures of about 2800 K and 45000 bar

being mandatory for melt growth will inhibit these substrates for the fore-

seeable future (Van Vechten 1973). All substrates other than GaN itself lead

to heteroepitaxial growth, thus giving rise to a deterioration in epitaxial

quality due to di€erences in lattice constants and thermal expansion coe-

cients between substrate and layer. For a comprehensive overview, potential

substrates for GaN technology are listed in Table 1 together with their

fundamental crystalline parameters.

Except of LiGaO

2

none of the potential substrates can provide lattice

matching even close to the requirements of other III±V technologies. In

Table 1. Potential substrates for GaN technology and their fundamental physical parameter

Substrate

Crystal structure

Lattice mismatch

to a-GaN (%)

at 300 K

Di€. in therm.

Expansion coe€.

to a-GaN (´10

)6

)

Cleavage

plane

Stability for

MOVPE

process

Si

Diamond

20.1

)2.0

(111)

Good

GaAs

Zincblende

25.3

0.4

(110)

Sucient

GaP

Zincblende

20.7

0.9

(110)

Sucient

MgO

Rocksalt

)6.5

4.9

(100)

Sucient

MnO

Rocksalt

)1.4

(100)

Instable

CoO

Rocksalt

)5.4

(100)

Instable

NiO

Rocksalt

)7.6

(100)

Instable

MgAl

2

O

4

Spinel

)10.3

1.9

(100)

Good

NdGaO

3

Perovskite

)1.2

1.9

Sucient

ZnO

Wurtzite

2.0

)2.7

(1±100)

Sucient

(11±20)

(0001)

6H-SiC

Zns 6H

)3.4

)1.4

(1±100)

Good

(11±20)

(0001)

LiAlO

2

b-NaFeO2

1.7

1.7

(001)

Instable

LiGaO

2

b-NaFeO2

)0.1

1.9

(010)

Instable

Al

2

O

3

Corundum

13.8

1.9

(1±102)

Good

LiNbO

3

Ilmenite

)6.7

9.9

(1±102)

Instable

LiTaO

3

Ilmenite

)6.8

10.6

(1±102)

Sucient

228

M. KAMP

addition to above properties, the thermal and the electrical conductivity as

well as price and availability have to be considered, leaving 6H-SiC and

c-plane sapphire (Al

2

O

3

) as the only two substrate widely used in GaN

technology.

Heteroepitaxial growth of GaN results in about 10

9

threading dislocations

(TD) per cm

2

for the present GaN technology on sapphire or SiC substrates.

Such high dislocation densities, being 5±6 orders of magnitude higher than in

conventional III±V technologies are present even in state of the art device

material. The deteriorating in¯uence of the TD, however, is signi®cant lower

than expected. Figure 1 shows the normalized eciency of LEDs versus the

dislocation density of these structures for a variety of di€erent III±V semi-

conductor systems including GaN (after Lester (1995)).

On the ®rst glance TD seem to be no problem to group III nitrides, as

judged from their LED eciency. However, as we will see later in Section 5,

`The Rule of Dislocations in GaN', TD severely hamper GaN based devices

in many terms including lifetime and performance.

3. Heteroepitaxial growth

Semiconductor technology requires epitaxial growth of an extremely high

quality. Perfect crystal growth can only be attained using a substrate that is

identical in crystal structure, lattice constant and thermal expansion coe-

cient. This is only guaranteed for homoepitaxy, where substrate and epitaxial

Fig. 1. Normalized eciency of LEDs versus dislocation density of these structures, for a variety of

di€erent III±V semiconductor systems.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

229

layer consist of identical material. Under those circumstances, layer-by-layer

growth can be obtained, resulting in two-dimensional growth without gen-

eration of dislocations. If a homoepitaxial substrate is not available, above

criteria should be matched as close as possible. Almost every semiconductor

material system either group IV, III±V or II±VI is grown using homoepitaxial

growth or at least a closely lattice matched substrate (Da=a  10

ÿ3

).

At growth temperature the overall mismatch between epitaxial ®lm and

substrate, resulting from di€erent thermal expansion coecients as well

as lattice mismatch at room temperature, is relaxed under formation of

dislocations. The lattice mismatch at growth temperature can be calculated

according to Equation (1)

Da

a

…T † ˆ

a

L

…T † ÿ a

S

…T †

a

S

…T †

ˆ

a

L

…20



C†…1 ‡ a

L

DT † ÿ a

S

…20



C†…1 ‡ a

S

DT †

a

S

…20



C†…1 ‡ a

S

DT †

…1†

where a

L

and a

S

are the temperature dependent lattice constants of layer and

substrate, respectively, a

L

and a

S

are their thermal expansion coecients, and

DT is the di€erence between room and growth temperature.

Upon cooling to room temperature di€erences in thermal expansion co-

ecients determine the residual biaxial stress in the epitaxial layer. The acting

forces (P), the stress (r) and the curvature radius (R) can be calculated

according to the two-dimensional elastic beam theory for isotropic materials.

P

i

ˆ Ed

i

P

j>i

d

j

ÿ

P

i

d

i

2R

‡

DT

d

X

j>i

d

j

…a

i

ÿ a

j

†

2
6

4

3
7

5

…2†

r

i

ˆ

P

i

d

i

‡

E
R

x

i

ÿ

d

i

2





…3†

R ˆ

…d

i

‡ d

j

†

3

6DT

P

i

P

j

d

i

d

j

…a

i

ÿ a

j

†

…4†

E being Youngs moduli, d

i

the thickness, and x

i

the distance as measured

from the central axis of the layer i.

The resulting strain, which can be up to 0.6 GPa for a 3 lm thick GaN

layer grown on sapphire account for serious macroscopic e€ects such a sig-

ni®cant curvature of the substrate/layer sandwich (Kozawa et al. 1995).

However, the key to heteroepitaxial growth is the stress release on the mi-

croscopic scale that can be approached considering the free energy of the

growing surface. Naturally and in general an ideal growing surface endeavors

230

M. KAMP

to minimize its free surface energy (c) which is usually achieved by a mini-

mization of the surface area, thus suppressing surface steps. However, in case

of a strained, and particularly of compressively strained, epitaxial ®lms it can

be energetically favorable to minimize the free energy by an undulation of the

surface (see Fig. 2).

Where amplitude t and period k of the roughness ful®ll the following

unequation (Pidduck et al. 1993):

t=k <

E…Da=a†

2

4cp

2

…5†

Under increasing and strong strain, however, the minimization of the surface

free energy eventually leads to the formation of a network of TD.

4. Generation of dislocations

The accommodation of lattice mis®t across the interface between an epitaxial

layer and its substrate was ®rst considered by Frank and van der Merwe

(1949). They showed that a mis®t smaller than about 7% can be accom-

modated by biaxial elastic strain until a critical thickness of the epitaxial ®lm

is reached. Above a certain strain, relaxation takes place by formation of

dislocations. For a given lattice mismatch, determined by the lattice pa-

rameters, thermal expansion coecients and growth temperature, the stress is

corresponding to a speci®c thickness, i.e., the so-called critical thickness (h

i

).

The concept evolved by Matthews and Blakeslee (1974) presumes that below

the critical thickness a dislocation-free, coherently strained interface is stable,

whereas a mis®t dislocation structure, being semicoherently strained, would

be stable for higher thickness.

Fig. 2. In¯uence of heteroepitaxially induced strain on the morphology and the formation of dislocations.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

231

The critical thickness can then be calculated by considering the forces

acting on a present TD. The force exerted on the given dislocation by the

mis®t strain is given by

F

misfit

ˆ

2G

1

…1 ‡ m†

…1 ÿ m†

fbl cos k cos b

…6†

The tension in the dislocation is given by

F

linetension

ˆ

G

1

G

2

p…1 ÿ m†…G

1

‡ G

2

†

b

2

…1 ÿ m cos

2

h†…ln…h=b† ‡ 1†

…7†

G being the bulk moduli, m the Poisson ratio, f the mis®t at the hetero

interface, b the length of the Burgers vector, l the length of the threading

segment, h the layer thickness. k being the angle between the Burgers vector

and the interfacial plane, b the angle between the normal of the slip plane and

the interfacial plane, h the angle between the Burgers vector and the line

direction of the mis®t segment.

If F

misfit

exceeds F

linetension

the dislocation will move within the interfacial

plane and form a mis®t dislocation, thereby destroying the coherence of the

interface.

The Matthews±Blakeslee model is will established and successfully applied

to most conventional III±V semiconductors. However, nitride semiconduc-

tors crystallizing in the wurtzite structure, reveal some peculiarities. The low

symmetry of the hexagonal system allows for multiple epitaxial orientations

being very similar but not identical in terms of their free surface energy and

chemical potential. The high c=a ratio of about 1.626, the narrow slip plane

spacing (d) and the length of the Burgers vector b have an direct impact on

the formation of mis®t TD according to the Matthews±Blakeslee model. As

compared to zincblende structures, the wurtzite have extraordinary high

Peierls forces (F

Peierls

) for c-type or screw dislocations (b ˆ ‡= ÿ ‰0001Š)

and mixed c,a-type dislocations (b ˆ 1=3 < 11±23 >), whereas the a-type or

edge dislocations (b ˆ 1=3 < 11±20 >) encounter only a small Peierls force

(Jahnen et al. 1998).

F

Peierls

ˆ 2blG

1

1 ÿ m cos

2

v

1 ÿ m

x exp ÿ2p

d…1 ÿ m cos

2

v†

b…1 ÿ m†

x





;

x ˆ exp

4
5

p

2

nk

B

T

G

1

V





…8†

d being the spacing of the slip planes and v the angle between the Burgers

vector and the line direction of the threading segment, n is the number

of atoms per unit cell, k

B

is Boltzmann's constant and T the growth tem-

perature.

232

M. KAMP

The Peierls force counteracts the driving shear force (F

misfit

) in addition to

the line tension force (F

linetension

) thus giving a new equilibrium condition for

the formation of mis®t TD.

F

misfit

ˆ F

linetension

‡ F

Peierls

…9†

For completion it should be noted that beside the Matthews±Blakeslee

model, a second model for strain relaxation was developed in 1963 by Van

der Merwe assuming that the interfacial energy between ®lm and substrate is

the minimum energy available for generation of mis®t dislocations (Van der

Merwe 1963). By minimizing the total energy, strain can be calculated as

function of layer thickness and the critical thickness is determined equating

the two energies.

5. The rule of dislocations in GaN

The generation of dislocations can hardly be avoided in lattice mismatched

material systems. Therefore, their repercussions have to be discussed.

Threading dislocations have strong in¯uences on the semiconductor material,

whereas most of the e€ects come along with serious limitations to device

performances, too.

 Threading dislocations are scattering centers for light propagating within

the crystal. They are therefore introducing losses, particularly deteriorating

laser performance. The in¯uence of threading dislocations has been inves-

tigated by Liau et al. (1996) who calculated an absorption of 3  10

2

cm

ÿ1

for a dislocation density of 2  10

10

cm

ÿ2

and decent assumptions for a

laser geometry. Using nowadays data, for a more reasonable estimate, one

would expect losses of about 1±10 cm

ÿ1

being in good agreement with

previously reported losses of about 45 cm

ÿ1

(Nakamura 1997).

 Screw type TD, potentially having on open core in the center, can create

nano-pipes with open diameters of 30±50 nm (Qian et al. 1995). Those

holes deteriorate the electrical properties of layers and devices by providing

low energy di€usion paths for contact metals, dopants and impurities

(Osinski et al. 1996). Since solid state di€usion is very low in GaN based

materials, di€usion along TD is supposed to be one of the major degra-

dation mechanisms for devices.

 TD also act as vertical shortcuts. Using TD `free' Epitaxial Lateral

Overgrown (ELOG) substrates, the reverse bias leakage current of

pn-junction diodes has been dramatically reduced by a factor of 1000

(Kozodoy et al. 1998).

 Threading dislocation can be regarded as charged line defects being re-

sponsible for the unexpected low mobilities observed in GaN technology.

TD are long known as scattering center in semiconductors. B. PodoÈr (1966)

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

233

calculated the impact of TD on the electron mobilities in Ge crystals as

early as 1966. He proposed the following dependence of the carrier mobility

(l) on the TD density (N), Debye length (L

D

) and temperature (T )

l /

…k

B

T †

3=2

L

D

N

…10†

This is in good agreement with experimental electron mobilities falling

signi®cantly short compared to mobilities above 1000 cm

2

/Vs as expected

from Monte±Carlo simulations. Recently Weimann et al. (1998) propose

that charged traps along dislocations, acting as scattering center for lateral

currents, are the dominating scattering mechanisms in highly deteriorated

GaN layers.

 Threading dislocations, being known for having a dislocation mobility 10

10

times lower than that of GaAs, gain extraordinary mobility with increasing

temperature (T  400



C). Thus for elevated temperatures, e.g. present in

high temperature electronics, TD become increasingly important for device

degradation (Sugiura 1997).

6. Concepts for dislocation reduction

As described earlier the stress induced by di€erent lattice constants and

thermal expansion coecients between layer and substrate, above a critical

thickness, is reduced by generation of dislocations. Fig. 3 shows TEM

micrographs of a MBE grown GaN layer deposited directly on Al

2

O

3

. The

pictures clearly reveal a multi-crystalline layer with several epitaxial orien-

tations not suitable for a device structure. Within III±V compound semi-

conductor technology various concepts have been developed and successfully

applied to overcome or reduce this limitation.

6.1.

STRAINED LAYER SUPERLATTICES

The introduction of strained layer superlattices (SLS) was, for example,

successfully used in GaAs/Si technology where the dislocation density could

be reduced from about 10

8

cm

ÿ2

to approx. 10

5

cm

ÿ2

. The general concept is

the bending of dislocations in the strain ®eld of the heterojunction. The TD

then can either follow the interface to the edge of the wafer or annihilate

themselves (see Fig. 4). However, in GaN technology there are no reports on

an ecient reduction in dislocation density by SLS. The low eciency in

dislocation reduction goes back to the high Peierls forces already discussed in

Section 4. The low gliding plane distance again inhibits the gliding of the

dislocations along the interface. However impurity gettering by SLS being

234

M. KAMP

known from many ®elds of III±V technology (Meier et al. 1994) is still an

issue for SLS in GaN technology.

6.2.

NUCLEATION LAYERS

Low temperature nucleation layers (NL) have initially been introduced into

GaN technology in 1986 by Amano et al. (1986). Thereby, for the ®rst time,

Fig. 3. TEM micrographs of a GaN layer grown directly on a sapphire surface.

Fig. 4. Concept of a Strain Layer Superlattice (SLS) structure for the reduction of dislocations.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

235

high quality GaN layers have been grown. The usage of NL provided a

breakthrough in GaN technology. Today a variety of various types of nu-

cleation layers is known from the literature, including GaN, AlN, GaN/AlN

combination layers, etc. Several groups report the usage of an additional

nitridation before NL growth, whereas other initiate growth directly after

degasing of the sapphire. The huge number of free parameters (i.e. V/III

ratio, temperature, thickness, growth rate, temperature ramps, crystallization

temperature, etc.) make the optimization of the NL extremely dicult and

time consuming. Additionally, these parameters seem to depend strongly and

non-linearly on each other. Optimized thickness and temperatures are sig-

ni®cantly di€erent for GaN or AlN NL and strongly depending on V/III

ratio, etc. Obviously, there is no converging into a particular set of param-

eters within the published data.

The one thing NL have in common is that they determine the defect

structure of the subsequently grown GaN layers and thereby strongly in¯u-

ence the quality of that material. The ways NL improves the GaN material

are as many and di€erent, as there are NL. They can for instance reduce the

residual strain of layers grown on sapphire substrates by new elastic strain

relaxation mechanisms (Albrecht et al. 1997). Figure 5 shows a GaN/Al

2

O

3

interface grown by MBE (Mayer et al., unpublished) depicting the selfaligned

periodic formation of grainlets with alternating compressive (13.8%) and

tensile (ÿ25.8%) strain where islands with di€erent orientations are com-

pressive and tensile strained.

Probably, the most important aspect with NL is that they provide nucle-

ation centers on the sapphire surface which form isolated island with facets

di€erent from (0001), such as the (01ÿ11) and the (01ÿ12) facet for instance

(Albrecht and Kamp, unpublished). If the growth rate of those facets is

higher than the (0001) growth rate, the islands will coalesce under formation

of low angle grain boundaries (see Fig. 6). This mechanism of preferential

lateral growth has indeed some similarities to the extremely successful ELOG

approach discussed later.

The impact of the initial stages of growth on the optical properties of a

2 lm thick GaN layer grown by GSMBE under identical condition is de-

picted in Fig. 7.

6.3.

NITRIDATION

Especially with sapphire substrates, one additional process step is often ap-

plied for further improvement of the NL. The bare sapphire surface is exposed

to the reactive nitrogen source at elevated temperatures. Depending on the

growth technique this can be activated atomic nitrogen in plasma enhanced

MBE (PEMBE) (Heinlein et al. 1997) or ammonia in reactive MBE (RMBE)

236

M. KAMP

(Grandjean et al. 1996) and MOVPE (Uchida et al. 1996). The intention is to

exchange oxygen atoms from the surface layers against the supplied nitrogen

atoms, thereby forming an Al

x

N

y

layer more suitable for epitaxial growth.

The time (degree) of nitridation has to be controlled carefully, since an

extended nitridation may eventually end in the formation of GaN whiskers.

The in¯uence of the nitridation on the density and kind of dislocations has

been elaborated by a careful investigation of J. Specks group (Wu et al.

1998). They report that a short nitridation (60 s) reduces the TD density from

1  10

10

to 2  10

8

cm

2

compared to a long nitridation of 400 s. Whereas the

short nitridation produces screw or mixed dislocations, the long nitridation is

reported to yield mainly pure edge dislocations. The material quality

achieved on the short nitridated layer is also signi®cantly improved.

The eciency of the nitridation is observed by X-ray photoelectron spec-

troscopy (XPS) and other techniques (Auger sputter pro®ling, re¯ection

high-energy electron di€raction, low energy electron di€raction) which re-

ported various degrees of eciencies for the nitridation. ECR plasma sources

are being more ecient than RF sources. Ammonia is found suitable for

nitridation in both RMBE and MOVPE.

Fig. 5. Sketch and corresponding TEM micrograph of a newly found strain release mechanism by a

periodic formation of grainlets with alternating compressive (13.8%) and tensile (ÿ25.8%) strain.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

237

Fig. 7. In¯uence of nitridation, nucleation and their combination on the PL (20 K) of a 2 lm GaN layer

deposited under otherwise identical conditions.

Fig. 6. Nucleation and coalescence of a nucleation layer. The established facets have an increased lateral

growth rate eventually yielding a closed and planar layer.

238

M. KAMP

Only very recently, it has been reported that the nitridation can yield a very

inhomogeneous surface. MOVPE overgrowth of those surfaces results into

GaN-layers with irregular Ga- and N-terminated areas that are deteriorating

the layer quality (Seelmann-Eggebert et al. 1997).

6.4.

MULTIPLE NUCLEATION LAYERS

The usage of repeated NL separeated by about 1 lm of high temperature

GaN layers was initially proposed by Amano et al. (1999). Figure 8 shows

the temperature and growth pro®le during the growth of a GaN layer using

multiple NL. Repeating such a layer sequence up to 7 NL, the TD density

can be reduced from initially 5  10

9

to 5  10

7

TD/cm

2

(Amano et al.

1999). The in¯uence of a second NL on the optical output power of an

InGaN/GaN MQW LED is shown in Fig. 9 (Schwegler et al. unpublished).

The second NL is grown identical to the initial one after deposition of 1 lm

GaN at high temperature (1050



C). A signi®cant increase in the optical

output power can be observed for the structure with two NL, indicating a

reduced density of non-radiative recombination centers.

6.5.

THICK LAYERS

Another concept successfully employed in GaN technology is the growth of

thick GaN layers, preferably by hydride vapor phase epitaxy (HVPE). As-

suming a certain probability for the termination and annihilation of dislo-

cations, the dislocation density can be reduced just by the growth of thick

layers. The reduction takes place by annihilation of dislocations and by local

abolition of the translation invariance of the crystal, by dislocations, strain

®elds, or even point defects (Beneking et al. 1985). Films up to 300 lm

thickness have been grown by HVPE. The epitaxial layers reveal a huge

Fig. 8. Scheme of a growth sequence for use of a multiple NL layer.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

239

vertical inhomogeneity with a strong reduction in carrier density and strain

within the ®rst 30 lm above the interface (Siegle et al. 1999). At the surface

the dislocation density is reduced down to 10

7

cm

ÿ2

and the free carrier

concentration can be as low as 1  10

17

cm

ÿ3

. Here, the lattice constants are

approximately the ones of GaN due to an almost complete relaxation. Once

diculties in crack formation and surface morphology are overcome by a

careful optimization of the growth process, high quality layers can be

achieved. Excellent PL data have been reported with clearly resolved A, B

and C free excitons and D



X linewidths as low as 0.8 meV (Meyer 1999). As

soon as those layers are commercially available, they are promising for use as

quasi-substrates in GaN technology.

6.6.

EPITAXIALLY LATERAL OVERGROWTH

As mentioned earlier, a low temperature nucleation layer grown under

appropriate conditions makes use of a lateral growth rate being signi®-

cantly higher than the vertical growth rate. The same physical e€ect is

used in ELOG GaN. This technique principally known from GaAs growth

on Si substrates was initially employed to GaN technology by Usui et al.

(1997).

First, a regular low temperature nucleation layer is deposited on sapphire

by MOVPE. Subsequently, an approx. 2 lm thick MOVPE GaN-layer is

deposited. The layer is then removed from the MOVPE system and about

0.1 lm thick SiO

2

or Si

x

N

y

masks are deposited on the surface preferably in

h1±100i direction. The width of the mask stripes is approx. 5 lm at a distance

of approximately 10 lm. After introduction of the masked layers into a

Fig. 9. In¯uence of a 2nd NL on the optical output power of an InGaN/GaN MQW LED.

240

M. KAMP

MOVPE or HVPE system, a GaN layer is epitaxially grown on top of the

structure to a thickness of about 20 or 200 lm, respectively.

As is indicated in Fig. 10 dislocations below the masked area cannot

propagate into the layer above. Only in the non-masked area, dislocations

will be able to continue into the upper layers. The lateral growth of the

subsequently deposited thick GaN layer leads to an overgrowth of the

masked area. Since the epitaxial information is from the sidewalls of the GaN

growing in the non-masked regions, the epitaxial quality is extremely high

with dislocation densities as low as 1  10

6

cm

2

and 3  10

7

cm

2

in masked

and windowed region, respectively. Since lateral overgrowth naturally takes

place from both sides of the mask, a single dislocation will occur in the

middle of the mask where both regions meet. In addition to the dislocation

generated at the concurrence of the low, but existing vertical growth rate,

lead to voids close to the center of the masks (Fig. 10).

The device quality can be signi®cantly improved as is shown by Nakamura

et al. (1999), who could increase LD cw-lifetime from about 50 to 10,000 h

introducing ELOG substrates and AlGaN/GaN modulation doped barriers.

Increasing the thickness of the ®nal GaN layer to about 200 lm, by

means of HVPE, allows for the separation of the GaN layer from the

sapphire substrate (Nakamura et al. 1998). The self-sustaining layers can

be separated by either polishing or by laser induced thermal dissociation in

a process similar to the one described in (Kelly et al. 1996). Laser diodes

fabricated on such freestanding GaN ®lms reveal a signi®cant increase in

lifetime. Comparing LD with an identical threshold current density,

devices on freestanding GaN ®lms have a reported lifetime four times

longer than their counterparts on sapphire (Nakamura, private commu-

nication). The improved performance is attributed to a reduced thermal

load of the devices and improved laser facets, which can be obtained by

simple cleaving, since device structure and quasi-substrate have the same

orientation.

Fig. 10. Schematics of the ELOG substrates depicting the processing steps and the obtained distribution

of the dislocation density.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

241

7. GaN homoepitaxy

Finally, the use of GaN single crystal substrates shall be discussed. As stated

earlier melt growth of GaN is impossible due to the extraordinary temper-

atures and pressures required for this process. However, the Polish High

Pressure Research Center (Unipress) succeeded in GaN growth by employing

a high pressure, high temperature process. GaN is formed from atomic

nitrogen dissolved in a Ga melt, a process requiring N

2

pressures of about

15 kbar and temperatures of about 1400



C (Porowski 1999). At a growth

rate of approximately 100 lm h

ÿ1

perpendicular to the c-plane, the wurtzite

crystals are grown up to areas of some 100 mm

2

at a thickness of about

200 lm. The crystal quality of the GaN substrates is excellent as indicated by

X-ray rocking curve measurements. Using CuK

a1

radiation, linewidths of 20

arcsec are obtained for the (0002) re¯ex. The excellent structural properties

are also pointed out by very low dislocation densities ranging from 10

3

±

10

5

cm

ÿ2

. The optical quality, however, is poor, near-bandgap excitonic

transitions are not visible, weak PL at 380 nm and at 530 nm is observable at

room temperature (RT).

Undoped crystals reveal a ¯at (000 ÿ 1) surface (i.e. N-polarity) and a

slightly rough (0001) surface (i.e. Ga-polarity). Both orientations of the un-

doped single crystal substrates have been investigated for growth under

identical conditions. Whereas the material quality of the (000 ÿ 1) surface is

still good (PL linewidth is approx. 5 meV) compared to heteroepitaxial

growth, the properties achieved on the (0001) surface are clearly superior.

The di€erences of the both orientations can be traced back to the di€erent

free surface energies of the orientations. From ab-initio calculations it is

determined that the free surface energy of the (000 ÿ 1) surface is signi®cantly

higher than the one of the (0001) surface (Zywietz et al. 1998). From this

point, the (0001) orientation provides a more stable surface with a lower

probability of dopant incorporation (Leszczynski and Meyer, unpublished).

Furthermore, both orientations have distinctly di€erent surface morpholo-

gies requiring a di€erent treatment. The almost ¯at (000 ÿ 1) surface can be

mechano-chemically polished to achieve an atomically ¯at surface, whereas

the rougher (0001) side is chemically inert and can be mechanically polished

only. The latter process leaves behind subsurface damage, which can be re-

moved by dry etching of about 300 nm. Fig. 11(a) shows a SEM micrograph

of a homoepitaxial grown GaN layer where only top half of the substrate was

dry etched before growth. The epitaxial layer on top of the dry etched part of

the substrate reveals an improved surface topography with almost no visible

scratches, trenches, or holes. Fig. 11(b) shows the corresponding CL intensity

distribution of the same region of the sample. On the etched part, the in-

tensity variation is almost negligible. In contrast, the are being not etched

yields only weak CL signals (1000 times less intensity) which also ¯uctuate

242

M. KAMP

locally. In addition to the improved intensity the pre-treated area of the

epitaxial layer shows a ten times narrower linewidth in CL (FWHM <

2 meV, still resolution limited) [Fig. 11(c)].

Homoepitaxial GaN layers with outstanding properties have been

achieved on (0001) surfaces using above described CAIBE technique. High

resolution PL at 4.2 K reveals free excitons A, B, C as well as excited states

of those excitons, where the identi®cation is veri®ed by re¯ectance

measurements also included in Fig. 12. The linewidth of the bound excep-

tions (3.464±3.472 eV) is as low as 0.1 meV.

Initial LED structures have been homotype pn-junction LEDs. Only

substrates that underwent a CAIBE treatment yield functional devices,

otherwise the metallization across the trenches causes a shortcut over the

pn-junction. The EL of homoepitaxial GaN pn-junction LEDs is depicted in

Fig. 13 for various current densities. The LEDs show an intense, single peak

emission at about 425 nm wavelength with a linewidth of 60 nm for low

Fig. 11. SEM image (11a), corresponding cathodoluminescence (CL) intensities (11b) and local

CL spectra (11c) obtained from an epitaxial GaN layer grown on partially CAIBE treated (0001) ±

oriented GaN substrates. CL measurements by F. Bertram, T. Riemann, and J. Christen, University

Magdeburg.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

243

currents. It is remarkable that the emission wavelength is at approx. 425 nm

even for current densities up to 3 kA cm

ÿ2

. As was initially pointed out by

Nakamura et al. (1991) this is a clear indicative for the high quality of the

p-type material obtainable by homoepitaxial growth. The EL obtained from

Fig. 12. Re¯ectance (above, linear scale) and low temperature photoluminescence (below, log. scale) of a

1.5 lm thick GaN layer grown by MOVPE on GaN single bulk substrates. The outstanding material

quality is express by the world record narrow linewidth and the observance of strong free exciton and their

excited states. The linewidth of the bound exciton is as narrow as 0.1 meV. Measurements by K. Kornitzer,

K. Thonke, and R. Sauer.

Fig. 13. Electroluminescence of a GaN homojunction pn-LED grown on GaN substrate. Emission

spectra are depicted at various current densities. At a given current density, the homoepitaxial devices are

twice as bright as comparable LEDs grown on sapphire (dashed line).

244

M. KAMP

heteroepitaxial LEDs grown on sapphire under identical conditions is also

included into Fig. 13 for comparison. The heteroepitaxial device reveals a

clear shift towards shorter wavelength being attributed to an inferior quality

of the p-material at the pn-junction. In addition, the data in Fig. 13 reveal

that the homoepitaxial LED is approximately twice as bright as their

counterpart on sapphire.

In addition to above homojunction LEDs, ®rst InGaN/GaN DH-LEDs

have been fabricated using single bulk crystal substrates (Kamp et al. 1999).

Compared to identical structures on sapphire substrates the homoepitaxial

heterostructure LEDs reveal an improved performance in particular at

low current densities, indicating a lower concentration of non-radiative

recombination centers. However, further work, depending on the avail-

ability of the substrates, has to be carried out to develop the homoepitaxial

device to a point where they become fully competitive to commercial LEDs.

8. Summary

The deteriorating in¯uence of threading dislocations in GaN is signi®cantly

smaller than in other semiconductor systems. However, with a dislocation

density being 6±7 orders of magnitude higher than with other III±V semi-

conductors, they still have a severe in¯uence on device performance and

lifetime. Within the paper, the origin of the formation of the dislocation has

been elucidated with special respect to GaN. Low temperature nucleation

layers can signi®cantly improve the material quality. Additional nucleation

Fig. 14. Electroluminescence of an InGaN/GaN LED grown on GaN single bulk crystal substrates.

Excellent EL is achieved at low current densities, revealing a low density of non-radiative defects.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

245

layers have been investigated and a further reduction of the dislocation

density is obtained using this moderate e€ort coming along with an in-

creasing LED device performance. Thick HVPE layer can also reduce the

dislocation density and in combination with ELOG can provide high quality

quasi-substrates successfully employed for GaN laser diodes. Bulk GaN

single crystal substrates, however, are the ultimate benchmark for GaN

technology. With this substrates PL linewidths as narrow as 0.1 meV have

been demonstrated (Fig. 14). First LED work on homojunction GaN and

heterojunction InGaN LEDs on GaN substrates is promising, with homo-

epitaxial LEDs being about twice as bright as their heteroepitaxial coun-

terparts. Being not di€erent from other semiconductor systems in that point,

growth of GaN on GaN (quasi-)substrates is clearly favorable over

heteroepitaxy.

Acknowledgements

The author is indebted and grateful to C. Kirchner, A. Pelzmann, M. Mayer,

V. Schwegler, and K.J. Ebeling from Department of Optoelectronics at

University of Ulm for their valuable contributions, continuous support and

helpful discussions. Without their work, this paper would never have been

written. Several other researchers contributed to this work, including K.

Kornitzer, K. Thonke, and R. Sauer from University Ulm, Department of

Semiconductor Physics (high resolution PL measurements and re¯ectance),

F. Bertram, T. Riemann, and J. Christen from University Magdeburg (CL

measurements), S. Christiansen, M. Albrecht, and H.P. Strunk from Uni-

versity Erlangen-NuÈrnberg (TEM measurements). The outstanding GaN

bulk crystal substrates have kindly been supplied by the Polish High Pressure

Research Center, namely by M. Leszcnynski, I. Grzegory, and S. Porowski.

The author gratefully acknowledge their valuable contributions. The GaN

project at the Department of Optoelectronics, in which most of the presented

experimental work was carried out, is partly funded by the German Federal

Ministry of Education, Science, Research and Technology (BMBF) and the

Volkswagen Foundation.

References

Albrecht, M., S. Christiansen and H.P. Strunk. Point strain sources compensating mis®t during epitaxial

growth. Appl. Phys. Lett. 70(8) 952±954, 1997.

Albrecht, M. and M. Kamp, unpublished.

Amano, H., N. Sawaki, I. Akasaki and Y. Toyoda. Metalorganic vapor phase epitaxial growth of a high

quality GaN ®lm using an AlN bu€er layer. Appl. Phys. Lett. 48 353±355, 1986.

Amano, H., M. Iwaya, N. Hayashi, T. Kashima, M. Katsuragawa, T. Takeuchi, C. Wetzel and I. Akasaki.

MRS Internet J. Nitride Semicond. Res. 4S1 G10.1, 1999.

246

M. KAMP

Lester, S.D., F.A. Ponce, M.G. Craford and D.A. Steigerwald. Appl. Phys. Lett. 66 1249 (1995).

Beneking, H., P. Narozny and N. Emeis. High quality epitaxial GaAs and InP wafers by isoelectronic

doping. Appl. Phys. Lett. 47 828±830, 1985.

Frank, F.C. and J.H. van der Merwe. Proc. R. Soc. London A198 216, 1949.

Grandjean, N., J. Massies and M. Lerous. Nitridation of sapphire. E€ect on the optical properties of GaN

epitaxial overlayers. Applied Physics Letters 69(14) 2071, 1996.

Heinlein, C., J.K. Grepstad, T. Berge and H. Riechert. Appl. Phys. Lett. 71 341, 1997.

Jahnen, B., M. Albrecht, W. Dorsch, S. Christiansen, H.P. Strunk, D. Hanser and Robert F. Davis, Pinholes,

Dislocations and Strain Relaxation in InGaN. MRS Internet J. Nitride Semicond. Res. 3 39, 1998.

Kamp M., C. Kirchner, V. Schwegler, A. Pelzmann, K.J. Ebeling, M. Leszczynski, I. Grzegory, T. Suski

and S. Porowski. GaN Homoepitaxy for Device Applications. MRS Internet J. Nitride Semicond. Res.

4S1 G10.2, 1999.

Kelly, M.K., O. Ambacher, B. Dahlheimer, G. Groos, R. Dimitrov, H. Angerer and M. Stutzmann.

Optical patterning of GaN ®lms. Appl. Phys. Lett. 69(12) 1749±1751, 1996.

Kozawa, T., T. Kachi, H. Kano, H. Nagase, N. Koide and K. Manabe. Thermal stress in GaN epitaxial

layers grown on sapphire substrates. J. Appl. Phys. 77(9) 4389±4392, 1995.

Kozodoy, P., J.P. Ibbetson, H. Marchand, P.T. Fini, S. Keller, J.S. Speck, S.P. DenBaars and U.K.

Mishra. Electrical characterization of GaN p-n junctions with and without threading dislocations. Appl.

Phys. Lett. 73(7) 975±977, 1998.

Leszczynski, M. and B.K. Meyer, unpublished.

Liau, Z.L., R.L. Aggarwal, P.A. Maki, R.J. Molnar, J.N. Walpole, R.C. Williamson and I. Melngallis.

Light scattering in high-dislocation- density GaN. Appl. Phys. Lett. 69(12) 1665±1667, 1996.

Matthews, J.W. and A.E. Blakeslee. J. Cryst. Growth 27 118, 1974.

Mayer, M., M. Kamp and M. Albrecht, unpublished.

Meier, H.P., M. Kamp and S. Strite. Role of Molecular Beam Epitaxy in the Field of Optoelectronics.

Microelectronics Journal. 25(8) 609±617, 1994.

Meyer, B.K. Free and bound exciton in GaN epitaxial ®lms. Mat. Res. Soc. Symp. Proc. 449 497±507,

1997.

Nakamura, S. Characteristics Of Room Temperature-CW Operated InGaN Multi-Quantum-Well-

Structure Laser Diodes. MRS Internet J. Nitride Semicond. Res. 2 5, 1997.

Nakamura, S. private communication.

Nakamura, S., T. Mukai and M. Senoh. Jpn. J. Appl. Phys. 30 (12A) L1998±L2001, 1991.

Nakamura S., M. Senoh, S. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, Y. Sugimoto,

T. Kozaki, H. Umemto, M. Sano and K. Chocho. Appl. Phys. Lett. 73 832, 1998.

Nakamura, S., M. Senoh, S. Nagahama, N. Iwasa, T. Matushita and T. Mukai. InGaN/GaN/AlGaN-

BASED LEDS and LASER DIODES. MRS Internet J. Nitride Semicond. Res. 4S1 G1.1, 1999.

Osinski, M., J. Zeller, P.C. Chiu, B.S. Phillips and D.L. Barton. AlGaN/InGaN/GaN blue light emitting

diode degradation under pulsed current stress. Appl. Phys. Lett. 69(7) 898±900, 1996.

Pidduck, A.J., D.J. Robbins and A.G. Gullis. In Microscopy of Semiconducting Materials A.G. Gullis,

J.L. Hutchison and A.E. Staton-Bevan eds, IOP Publishing, Bristol 1993.

PoÈdor, B. Phys. Stat. Solidi, 16 K167, 1966.

Porowski, S. Near defect free GaN substrates. MRS Internet J. Nitride Semicond. Res. 4S1 G1.3, 1999.

Qian, W., G.S. Rohrer, M.S. Skowronski, K. Doverspike, L.B. Rowland and D.K. Gaskill. Open-core

screw dislocations in GaN epilayers observed by scanning force microscopy and high-resolution

transmission electron microscopy. Appl. Phys. Lett. 67(16) 2284±2286, 1995.

Schauler, M., F. Eberhard, C. Kirchner, V. Schwegler, A. Pelzmann, M. Kamp, K.J. Ebeling, F. Bertram,

T. Riemann, J. Christen, M. Leszczynski, I. Grzegory, T. Suski and S. Porowski. Dry etching of GaN

substrates for high-quality homoepitaxy. Appl. Phys. Lett. 74(8) 1123± 1125, 1999.

Schwegler, V., C. Kirchner and M. Kamp, unpublished.

Seelmann-Eggebert, M., H. Zimmermann, H. Obloh, R. Niebuhr and B. Wachtendorf. Plasma cleaning

und nitridation of sapphire substrates for AlGaN epitaxy as studied by ARXPS and XPD. Mat. Res.

Soc. Symp. Proc. 468 193, 1997.

Siegle, H., A. Ho€mann, L. Eckey, C. Thomsen, J. Christen, F. Bertram, D. Schmidt, D. Rudlo€ and

K. Hirmatsu. Vertical strain and doping gradients in thick GaN layers. Appl. Phys. Lett. 71(17) 2490±

2492, 1999.

SOLUTIONS FOR HETEROEPITAXIAL GROWTH OF GaN

247

Sugiura, L. Dislocation motion in GaN light-emitting devices and its e€ect on device lifetime. J. Appl.

Phys. 81(4) 1633±1638, 1997.

Uchida, K., A. Watanabe, F. Yano, M. Kouguchi, T. Tanaka and S. Minagawa. Nitridataion process of

sapphire substrate surface and its e€ect on the growth of GaN. J. Appl. Phys. 79 7, 1996.

Usui, A., H. Sunakawa, A. Sakai and A.A. Yamagucki. Jpn. J. Appl. Phys. 36 L899±L901, 1997.

van der Merwe, J.H. J. Appl. Phys. 34 117, 1963.

Van Vechten, J.A. Phys. Rev. B7 9, 1973.

Weimann, N., L.F. Eastman, D. Doppalapudi, H.M. Ng and T.D. Moustakes. Scattering of electrons at

threading dislocations in GaN. Appl. Phys. Lett. 83(7) 3656±3658, 1998.

Wu, X.H., P. Fini, E.J. Tarsa, B. Heying, S. Keller, U.K. Mishra, S.P. DenBaars and J.S. Speck. Dis-

location generation in GaN heteroepitaxy. Journal of Crystal Growth 189/190 231±243, 1998.

Zywietz, T., J. Neugebauer, M. Sche‚er, J. Northrup and Chris G. Van de Walle. MRS Internet J. Nitride

Semicond. Res. 3 26, 1998.

248

M. KAMP