Effect of Si doping on strain, cracking, and microstructure in GaN thin films grown by metalorganic chemical vapor deposition

L. T. Romano, C. G. Van de Walle, J. W. Ager, W. Gtz, and R. S. Kern Citation: J. Appl. Phys. 87, 7745 (2000); doi: 10.1063/1.373529 View online: http://dx.doi.org/10.1063/1.373529 View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v87/i11 Published by the American Institute of Physics.

Related Articles
Why does the second peak of pair correlation functions split in quasi-two-dimensional disordered films? Appl. Phys. Lett. 102, 071907 (2013) Low-bias electron transport properties of germanium telluride ultrathin films J. Appl. Phys. 113, 063711 (2013) P(VDF-TrFE-CFE) terpolymer thin-film for high performance nonvolatile memory Appl. Phys. Lett. 102, 063103 (2013) Investigation of the near-surface structures of polar InN films by chemical-state-discriminated hard X-ray photoelectron diffraction Appl. Phys. Lett. 102, 031914 (2013) Effects of grain microstructure on magnetic properties in FePtAg-C media for heat assisted magnetic recording J. Appl. Phys. 113, 043910 (2013)

Additional information on J. Appl. Phys.

Journal Homepage: http://jap.aip.org/ Journal Information: http://jap.aip.org/about/about_the_journal Top downloads: http://jap.aip.org/features/most_downloaded Information for Authors: http://jap.aip.org/authors

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

JOURNAL OF APPLIED PHYSICS

VOLUME 87, NUMBER 11

1 JUNE 2000

Effect of Si doping on strain, cracking, and microstructure in GaN thin lms grown by metalorganic chemical vapor deposition
L. T. Romanoa) and C. G. Van de Walle
Xerox Palo Alto Research Center, Palo Alto, California 94304

J. W. Ager III
Materials Sciences Division, Lawrence Berkeley National Laboratory, California 94720

W. Gotz and R. S. Kern

LumiLeds Lighting, San Jose, California 95131

Received 20 December 1999; accepted for publication 14 February 2000 The effect of Si doping on the strain and microstructure in GaN lms grown on sapphire by metalorganic chemical vapor deposition was investigated. Strain was measured quantitatively by x-ray diffraction, Raman spectroscopy, and wafer curvature techniques. It was found that for a Si concentration of 2 1019 cm 3 , the threshold for crack formation during lm growth was 2.0 m. Transmission electron microscopy and micro-Raman observations showed that cracking proceeds without plastic deformation i.e., dislocation motion , and occurs catastrophically along the low energy 1100 cleavage plane of GaN. First-principles calculations were used to show that the substitution of Si for Ga in the lattice causes only negligible changes in the lattice constant. The cracking is attributed to tensile stress in the lm present at the growth temperature. The increase in tensile stress caused by Si doping is discussed in terms of a crystallite coalescence model. 2000 American Institute of Physics. S0021-8979 00 03410-1

I. INTRODUCTION

Growth of IIIV nitrides for device applications is primarily done on sapphire substrates using metalorganic chemical vapor deposition MOCVD .1 Since the thermal expansion coefcient of sapphire is larger than that of GaN, a compressive stress estimated at greater than 1 GPa would be expected to develop in a lm that was grown stress free at the growth temperature typically 1000 C .2 Typically, smaller, but still compressive stresses are observed.26 An interesting effect has been observed with Si doping. Highly Si doped GaN is desired to form low resistance n-type conductive layers in nitride-based light emitting diodes, lasers, and optoelectronic devices. However, the maximum amount of Si doping appears to be limited by cracks that develop in lms with high Si concentration.7,8 Ex situ Raman and x-ray diffraction XRD studies performed on Si doped GaN lms have shown that the compressive strain observed at room temperature in MOCVD-grown lms decreases with increasing Si concentration.6 It has also been suggested that the reduction of compressive stress observed at room temperature in moderately Si doped samples ( 3 1018 cm 3 ) is due to the formation of dislocations in the basal plane.5 Recently, in situ wafer curvature measurements have shown that undoped GaN lms are in tension at the growth temperature.9 We perform here a detailed study of the effect of Si doping on the strain and microstructure of MOCVD-grown GaN lms. First-principles calculations were used to determine the change in lattice constant expected for Si substituta

ing for Ga in GaN. Film strain was measured with three complementary techniques: XRD, wafer curvature, and micro-Raman spectroscopy with 1 m lateral resolution. The microstructure was evaluated with transmission electron microscopy TEM , scanning electron microscopy SEM , and atomic force microscopy AFM . Strain gradients in the vicinity of cracks were measured with micro-Raman spectroscopy. The experimental observations are discussed in terms of a crystallite or domain coalescence model developed for the growth of polycrystalline lms.10
II. EXPERIMENT A. Samples

Electronic mail: romano@parc.xerox 7745

GaN lms were grown in conventional MOCVD reactors on 0001 sapphire substrates similar to studies described elsewhere using trimethylgallium, ammonia, and hydrogen as a carrier gas.11 Silane was introduced into the gas stream to achieve Si doping in the range from 3 1016 to 5 1019 cm 3 . The standard deposition procedure for Sidoped lms A lms was sequential growth of: 1 a low temperature LT nucleation layer, 2 a 100 nm high temperature HT undoped layer, and 3 12.5 m HT Si-doped layer. Some samples B lms were made without the undoped HT layer i.e., silane was introduced immediately after growth of the LT nucleation layer . The total lm thickness h was varied between 1 and 3.0 m. Hall-effect and secondary ion mass spectrometry measurements were used to determine the electron and Si concentrations Si . In all the samples the quantitative agreement between Si and the carrier concentration indicated that all the Si was electrically active and it was therefore inferred to
2000 American Institute of Physics

0021-8979/2000/87(11)/7745/8/$17.00

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

7746

J. Appl. Phys., Vol. 87, No. 11, 1 June 2000

Romano et al.

reside on the Ga site. The surface roughness and microstructure of the lms were measured by a combination of AFM, SEM, and TEM. Samples for TEM were prepared in the standard way by mechanically thinning and ion milling to electron transparency.

B. Strain determination from x-ray diffraction

The a and c lattice constants were measured by XRD using the (1015) and 0002 reections, respectively. The strain values can then be obtained from the equations
a

a/a 0 1

1
FIG. 1. a and c lattice constants measured by x-ray diffraction for 2 m thick GaN lms as a function of Si . Inverted triangle symbols are A lms with undoped interlayer , diamond symbols are B lms no undoped interlayer . Hatched open symbols are the a lattice constants, solid symbols are the c lattice constants. a sc and c sc are the values for the bulk lattice constants from Ref. 12 . The solid line through the points is a guide and not meant to imply the transition is continuous.

and
c

c/c 0 1.

To calculate these strains, the lattice constants a 0 and c 0 of the relaxed, unstrained bulk material must be known. Among the reported values, it is our opinion that the ones that most closely approximate the true lattice constant of pure GaN are the values measured for undoped homoepitaxial layers grown on top of bulk single crystals with values of a 0 3.1878 and c 0 5.1850 . 12 These values are extremely close to the values of a and c measured for bulk single crystal GaN grown under high pressure in the presence of Mg (a 0 3.1876 and c 0 5.1846 .13

D. Strain determination from curvature measurements

Wafer curvature measurements of the lms were performed in a Tencor laser system using a sapphire substrate as a reference. The curvature is equal to the inverse of the radius of curvature. The stress in the lm, a , is related to the curvature by Stoneys Equation16 see Ref. 9
ah 2 M s d Al2O3 /6,

C. Strain determination from Raman measurements

In-plane strain ( a/a a ) was measured with a lateral spatial resolution of 1 m with micro-Raman spectroscopy using shifts of the E2 phonon. The E2 phonon line in GaN is sensitive to strain in the layer, and its frequency can be accurately measured using Raman spectroscopy. In Ref. 14, the following relationship was derived between the measured shift of the phonon line E2 and the stress in the layer: E2 The stress sion
a a a

where h is the thickness of the GaN lm, d Al2O3 is the substrate thickness 430 m in our case , and M s is the biaxial modulus of the substrate see Eq. 5 . We have used the value M s 610 GPa for Al2O3 , as derived from the elastic constants in Ref. 17. Equation 7 thus allows us to determine the stress in the lm, and the strain can subsequently be derived using Eq. 4 .
III. RESULTS A. Strain

2.4 0.2 cm 1/GPa . is related to the strain ,

a

through the expres4

a /M f

where M is the biaxial modulus of GaN, which is dened as Mf

2 c 11 c 12 2c 13/c 33 .

Elastic constants for GaN were taken from Ref. 15, yielding a value for M 478 GPa. As discussed in Ref. 14, the Raman measurements produce an accurate assessment of the slope of the stress versus phonon shift curve, but the absolute magnitude of stresses and strains depend on the choice of a reference lattice constant. The standard used in the Raman shift measurements a bulk single crystal has lattice constants slightly larger than the homoepitaxial lms which we choose as our reference, as discussed above. The relationship between the measured Raman shift and strain is then
a

0.000 62 0.000 86 E2 .

The a and c lattice constants were measured for all the lms by XRD. The results for the 2 m lms are shown in Fig. 1 for samples A undoped prelayer and B no undoped prelayer . The a lattice constants are found to be smaller and c lattice constants larger than the value found for bulk single crystals for lms with Si 1018 cm 3 , whereas the a lattice constant is greater and c lattice constants smaller than the bulk values for Si 1018 cm 3 . The values for the single crystal lattice parameters a sc and c sc are noted in Fig. 1 from Ref. 12. Figure 2 a is a combined plot of the in-plane strain a measured by XRD, Raman, and curvature techniques as a function of Si doping for 2 m thick, A and B lms. The strains were calculated as discussed in Sec. II. The gure shows that our three different techniques for measuring the strain produce consistent results. Some exceptions are noted for lms at the highest Si , where cracks were present and

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

J. Appl. Phys., Vol. 87, No. 11, 1 June 2000

Romano et al.

7747

B. Extended defect structure observed by TEM

FIG. 2. a Comparison of in-plane strain a measured for 2 m thick GaN lms as a function of Si by XRD squares , micro-Raman spectroscopy triangles , and wafer curvature circles . The circles with a cross are A lms and the open circles are B lms. The trend of increasing tensile strain with increasing Si is observed by all three techniques. The solid line through the points is a guide and not meant to imply the transition is continuous. b Comparison of in-plane strain a as a function of Si for 1 and 2 m thick lms as measured by micro-Raman. Solid circles are 1 m thick, A lms; circles with a cross are 2 m thick, A lms; open circles are 2 m thick, B lms. The solid line through the points is a guide and not meant to imply the transition is continuous.

found to cause local strain variations. XRD and wafer curvature measure strain over a macroscopic area, whereas the Raman microprobe is able to measure local strain variations. Strain variations around cracks will be discussed in Sec. III C. Figure 2 a shows that the in-plane strain increases with Si and the values of the strain for A and B lms are comparable in magnitude. At low Si the lms are in compression, which is expected to occur on cooling, due to the difference in thermal expansion coefcients between GaN and sapphire. However, as the Si concentration increases, the lms are found to be in in-plane tension. The transition from compressive to tensile strain occurs at a Si between 1018 1019 cm 3 . Figure 2 b shows a comparison of the strain for the 1 and 2 m thick lms in which only the Raman measurements are shown for simplicity. The 1 m thick lms clearly exhibit a lower strain than the 2 m thick lms for doping concentrations exceeding 1018 cm 3 .

The defect structure was investigated by TEM for A and B samples with different Si . It was found that the density and the type of dislocations in the lms were dependent on the doping level and presence of the undoped prelayer. The TEM images shown in Fig. 3 are for lms grown at a doping concentration and thickness for which cracking was not observed. Figures 3 a and 3 b were obtained using a diffraction vector g-5g g 0002 weak beam condition, at a tilt approximately ten degrees from the 1120 towards the 1010 zone. Figure 3 a is for a 1 m thick lm with Si 2 1019 cm 3 grown on a 100 nm undoped prelayer. The image shows a broad band of contrast near the interface between the undoped prelayer and the doped overlayer. This contrast is due to the interference of overlapping regions with different c and a lattice constants across the doped/ undoped interface as measured by XRD.18 The broad fringe contrast was not observed for a 1 m thick lm with Si 5 1018 cm 3 as shown in Fig. 3 b , where there was no signicant lattice constant change between the undoped and doped layer. The threading dislocations in both of the images either have mixed or screw character with a component of the burgers vector b along the 0001 direction. Dislocations with burgers vector b parallel to the 1120 direction are imaged in Fig. 3 c (g 1010) for the same region of lm shown in Fig. 3 a . The combination of both images in Fig. 3 a and 3 c show that the threading dislocations are unperturbed at the boundary between the undoped and doped layers and no additional dislocations are created. In addition, no dislocations along the basal plane were observed. This observation is consistent with the difculty of generating mist dislocations on the basal plane at lattice mismatched interfaces found previously for thick InGaN layers on GaN19 and thin lms with high In content in quantum well structures.20 By comparison a 3.0 m thick lm grown without an undoped prelayer for Si 5 1018 cm 3 is shown in Fig. 3 d using diffraction conditions g 1010. It can be observed that predominately basal plane dislocations are found within a 1 m thick region above the LT buffer layer. The presence of dislocations on the basal planes increases the probability of defect interactions with threading dislocations. This may explain the decrease in dislocation density at the lm surface with increasing Si observed by other groups.5 However, for the lms in our study no such trend was observed for lms grown with an undoped prelayer since basal plane dislocations were typically not present. As discussed above, the dislocations that were generated in the undoped prelayer extended unperturbed to the surface resulting in a defect density at the surface of 1010/cm2 for Si up to 2 1019 cm 3 .
C. Cracking

Cracking was observed for both A and B lms with a doping level of 2 1019 cm 3 for h 2 m. The cracks were found to either extend partially (h 2 m) or completely from the lm/substrate interface to the surface (h 2.5 m). The SEM micrographs in Fig. 4 for an A lm that was 2 m thick, shows that the crack occurs on the

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

7748

J. Appl. Phys., Vol. 87, No. 11, 1 June 2000

Romano et al.

FIG. 3. TEM images for 1 m thick, A lms with Si 2 1019 cm 3 in a and c ; and Si 5 1018 cm 3 in b . The images in a and b were using a diffraction vector g-5g (g 0002) weak beam conditions. A 3.0 m thick B lm with Si 5 1018 cm 3 is shown in d . The images in c and d are dark eld images with g 1010.

prismatic 1100 cleavage planes21 of the GaN i.e., parallel to the growth direction and extends into the substrate. Figures 5 a 5 c are TEM micrographs of a subsurface crack in the same lm. Dislocation images taken in the same region along g 1010 Fig. 5 a and g 0002 Fig. 5 b near the 1120 zone, show that no additional dislocations are generated near the crack. This suggests that plastic deformation does not take place to promote crack formation. The widest part of the crack is revealed using the imaging conditions in Fig. 5 b to be L 50 nm and the average crack spacing L is shown in Fig. 5 c to be 10 m. As shown in the AFM image and corresponding line scan in Fig. 6, the surface of the lm that grows over the cracks is higher than the region between the cracks. The lighter contrast lines are the position of the cracks below the lm surface indicating the crack pattern and a crack spacing of 10 m for parallel cracks, consistent with the TEM results. For the position measured in Fig. 6, the surface of the overgrown lm was 16 nm higher above the crack compared to the region between the cracks. Typical values for the root mean square roughness

between the cracks was found to be 3 nm whereas the surface height over the cracks typically ranged between 1020 nm. The subsurface cracking suggests that the lms cracked during growth at high temperature, which is consistent with our measurements showing tensile strain in these lms. Micro-Raman measurements of strains around a crack in a 2.5 m thick A lm with Si 2 1019 cm 3 are shown in Fig. 7. The inset in Fig. 7 is a SEM showing that the crack extends from the lm substrate to the lm surface. At distances far away from the crack 75 m , the in-plane strain is 0.015 tensile . The strain decreases near the crack, and parts of the lm are in a small amount of compression 0.005 at the crack location. We attribute the compressive strain at the crack to thermal stresses occurring on lm cool down we assume the cracking occurred at the growth temperature . Several groups2224 have studied the elastic stress eld around surface cracks in thin lms. We apply here the model in Ref. 23, assuming that by symmetry, a surface crack in a stressed lm is equivalent to a semi-innite

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

J. Appl. Phys., Vol. 87, No. 11, 1 June 2000

Romano et al.

7749

FIG. 4. SEM micrographs of a crack in a 2 m thick, A lm with Si 2.0 1019 cm 3 . a The crack extends to sapphire substrate and b extends along the 1100 cleavage plane of the GaN.

edge.22 The adjustable parameters used in the model include the elastic constants and thickness of the lm and substrate, and the value of the lm stress away from the crack. The elastic constants used for GaN were taken from measurements by Polian et al.25 and the sapphire elastic constants were from Tefft.17 A lm thickness of two microns was obtained from the SEM measurement and the in-plane strain away from the crack was taken from our micro-Raman measurement to be 0.0015 equivalent to a tensile biaxial stress of 560 MPa . As shown in Fig. 7, there is good agreement of the elastic cracking model solid line and the values of strain measured by micro-Raman. Therefore, we conclude that cracking of the lm leads to elastic relaxation in the vicinity of the crack which is consistent with the lack of dislocations observed near the cracks Fig. 5 . This is in contrast with the well-studied case of stress gradients in SiO2 /Si devices, where lm stresses have been observed to lead to both dislocation generation and cracking.26

FIG. 6. a AFM image of the surface of the lm shown in Fig. 5 and b corresponding line scans associated with the overgrown subsurface cracks.

IV. FIRST-PRINCIPLES CALCULATIONS

The observed dependence of strain on Si concentration could, in principle, be due to a change in lattice constant induced by Si incorporation. Incorporation of dopant impurities has two distinct effects on the lattice constant.27 The rst effect is purely a size effect, and is related to the difference in atomic radius between the impurity and the host atom that is replaced by the Si. The second effect is an electronic effect, related to deformation potentials. In the case of n-type doping, the donors contribute electrons to the conduction band. The energy of the system can therefore be lowered if the conduction band energy can be decreased on an absolute energy scale . Such a shift in the position of the conduction band can occur if the volume of the material is changed; hydrostatic strain indeed leads to a change in the conduction

FIG. 5. TEM micrographs of a subsurface cracking in a GaN lm h 2.0 m, Si 2 1019 cm 3 , no undoped interlayer : High magnication image showing crack that is overgrown and extends into sapphire substrate with diffraction vector a g 0002 and b g 1010. Lower magnication image in c showing distance between two cracks.

FIG. 7. Micro-Raman measurements of in-plane strain across a crack cf. inset in a GaN lm grown on sapphire h 2.5 m, Si 2 1019 cm 3 , no undoped interlayer . The solid line is the expected elastic strain relaxation predicted by the model of Atkinson et al., Ref. 23.

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

7750

J. Appl. Phys., Vol. 87, No. 11, 1 June 2000

Romano et al.

band energy, with a proportionality factor given by the deformation potential. Such strain also bears a cost, of course, because of the elastic energy associated with the deformation. This elastic energy, however, varies as the square of the strain, whereas the energy gain due to the deformation varies linearly with the strain; the system can therefore lower its energy by undergoing a deformation. The size effect can be addressed by studying the relaxation of host atoms around the impurity. Even though we are dealing with wurtzite semiconductors, we treat the material as isotropic. Our results indeed indicate that the anisotropy of the atomic relaxations is very small compared to the magnitude of the overall relaxations. It is customary to dene a parameter size which relates the fractional change in the lattice constant to the impurity concentration a/a
size

We thus obtain e 1.5 10 24 cm3 , the positive sign indicating that placing electrons in the conduction band results in an expansion of the lattice. We conclude that for Si the size effect and deformationpotential effects have similar magnitudes but are of opposite sign, resulting in a very small net strain effect for SiGa. Assuming complete ionization, we can write a/a tot Si , with tot size 0.1 10 24 cm3 . For Si 2 1019 e cm 3 the maximum concentration used in this work the expected in-plane strain is 2 10 6 . This value is much smaller than the range of strain values observed as a function of Si 1.1 10 3 2.5 10 3 , as reported above . Therefore, we conclude that the changes in lattice constant as a function of Si cannot be attributed to lattice distortions caused by the substitution.

Si].

8
V. DISCUSSION A. Effect of silicon on the lattice constant

The calculations were performed using density-functional theory in the local-density approximation, ab initio pseudopotentials, and a supercell geometry. This is the same computational approach that we have successfully applied to the study of the atomic and electronic structure of a wide variety of defects and impurities in nitride semiconductors.28,29 The calculations show that the nitrogen atoms surrounding the Si donor relax inwards by about 0.10 around the SiGa in the positive charge state . In order to determine the change in lattice constant, we follow a procedure described in Ref. 27. The idea is to use Vegards law i.e., a linear interpolation to estimate the contribution of atom-size differences to the observed lattice parameter. The end-point structures for the interpolation are pure GaN 0% Si and a hypothetical zincblend SiN compound. Following this approach, we obtain a 1.6 10 24 cm3 . The negative sign of size invalue size dicates that the incorporation of Si atoms to the lattice results in a decrease of the lattice constant. The deformation-potential effect can be derived by performing a minimization of the sum of the elastic energy cost quadratic in the strain and the energy gained due to lowering of the conduction band, occupied with a certain number of electrons linear in the strain . Similar to the parameter size dened in Eq. 8 , one can dene a parameter e, where the subscript indicates the deformation-potential effect for electrons a/a
en

In Sec. IV we argued, based on rst-principles calculations, that incorporation of silicon does not affect the lattice constant of the GaN overlayer. This result can actually be conrmed purely based on the measured lattice constants by invoking the theory of elasticity. From the theory of elasticity we know that the strains are related by the expression:
c

a,

11

and n is the electron concentration. Note that the carrier concentration n is not necessarily equal to the donor concentration, because of compensation or incomplete ionization. However, as discussed in Sec. II A, the samples studied in the present work show complete electrical activation of the Si donors and since most of the donors are ionized at room temperature, we can assume n Si . The shift of the conduction band minimum is described by the absolute deformation potential a c , as dened in Ref. 27. e can be calculated as
e

a c /3B,

10

where B is the bulk modulus of GaN. The absolute deforma6.0 eV). tion potential a c was calculated in Ref. 30 (a c

where D 2c 13 /c 33 . Theoretical15 and experimental25 determinations of the elastic constants have produced a consistent and reliable value for D, namely D 0.51. This corresponds 0.20, which is also well to a value of Poissons ratio accepted.31 Here we systematically use the elastic constants derived by Wright,15 which are very close to the experimental values of Polian et al.25 Combining Eqs. 1 , 2 , and 13 , we can use a given pair of measured a and c values for a strained layer to derive the unstrained lattice constants of that layer, by assuming the above value of D, and a particular value for the c/a ratio; for the latter, we have chosen 1.6265. For all of the layers examined in the present study, the a 0 lattice constant extracted from this analysis is within 0.0010 of the value a 0 3.1878 quoted above for pure GaN with a systematic deviation towards larger values . This deviation is very small, when compared to lattice-constant variations reported in the literature see, e.g., Ref. 32 . This result conrms our conclusion that the effect of silicon incorporation on the lattice constant of our layers is very small. Figure 8 shows c versus a for all the 1 and 2 m lms in our study. The solid line corresponds to the prediction from the theory of elasticity, for a biaxially strained lm with given bulk lattice constants a 0 3.1878 and c 0 5.1850 . The deviation from the solid line is smaller than the measured variation in a and c see Fig. 1 . This indicates that the measured values are due to variations in the strain induced on the lm rather than a change in the bulk lattice constants.

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

J. Appl. Phys., Vol. 87, No. 11, 1 June 2000

Romano et al.

7751

FIG. 8. Lattice constants c and a in angstrom for all the 1 and 2 m lms examined in the present study. The solid line is the prediction from elasticity theory for strained lms with lattice constants a 0 3.1878 and c 0 5.1850 . A different choice of bulk lattice constants would produce a line parallel to the one shown in the gure.

lms showed no evidence of cracks. It is possible that gaps on the surface, due to the roughness, coalesce in a similar way as the crystallites described in the model by Nix and Clemens NC . Limited surface diffusion of atoms on the growing surface would cause adatoms to attach to the strained crystal at their point of arrival, and the lm would grow in a strained state. If we associate the observed lm roughness with an effective crystallite size, then as the roughness increased the effective crystallite would decrease. According to the NC model, this would predict a higher tensile stress with increasing roughness high Si . The cause for the increased surface roughness with increasing Si is not understood. The presence of Si may limit the surface diffusion of the Ga and N atoms on the growing surface. The presence of Si on the surfaces of AlGaN35 and InGaN36 has been shown to lead to GaN quantum dot formation35 and spiral growth.36 The competition for N atoms to form SiN may also play a role in the limited surface diffusion of the Ga and N atoms. It has been found that the growth rate of SiN grown by MOCVD was found to be three times higher when increasing the temperature from 800 to 1080 C.37,38
VI. CONCLUSIONS

B. Dependence of strain on silicon concentration

The increase in tensile strain with Si doping is puzzling. We have shown above that incorporating Si in the GaN lattice substituting on a Ga site has a negligible effect on the lattice constant of the GaN lattice. Silicon must therefore play a different role in introducing strain in the material. Recently Hearne et al.9 reported in situ wafer curvature measurements that show that tensile strains occur in GaN lms at the growth temperature. The lms in their study were nominally undoped, and exhibited tensile stresses up to 0.6 GPa at the growth temperature. At room temperature the lms in the study by Hearne et al. were found to be in compression, which is consistent with our results for lms with Si 1018 cm 3 . Thermal cycling between 1000 and 450 C showed that plastic relaxation of the growth stress does not occur during the cool down.9 This is consistent with our results on lms that were cracked without creating additional dislocations. Except for cracking, there was no obvious effect of Si on the microstructure for lms grown with the undoped prelayer. Furthermore, the cracks generated during the growth occurred without the creation of dislocations. One possible explanation of the lm stress present during growth may be as a consequence of lm coalescence. The intrinsic stresses that develop during the growth of polycrystalline lms have been investigated for many years.33,34 Tensile stress is often found in these lms and attributed to attractive forces between the grains. Nix and Clemens10 have recently developed a quantitative model for this process. In this model, tensile lm stress increases monotonically with decreasing crystallite size. In our study, the GaN lms are single crystal. However, the surface structure, as measured by AFM, was found to depend on the Si concentration. We measured the surface roughness of two Si doped lms with a thickness of 20 nm grown on undoped prelayers by AFM. Increasing Si from 5 1017 to 5 1019 cm 3 leads to a doubling of the root mean square surface roughness from 5.5 to 11.0 nm . These

Crack formation in GaN lms doped with Si was studied by measuring the strain and comparing to the microstructure. The strain was found to increase with lm thickness and Si concentration. Crack formation along the 1100 cleavage planes was found to be the major strain relief mechanism and occurred without creating additional dislocations. This suggests that plastic deformation is difcult in these materials at the temperatures used during growth. First-principles calculations showed that silicon incorporation does not affect the lattice constant of the GaN crystal. It was proposed that the increase in tensile stress with increasing Si is related to the presence of tensile stress due to crystallite coalescence, in accordance with a model established recently by Nix and Clemens.10
ACKNOWLEDGMENTS

We acknowledge helpful discussions with John Northrup, Stacia Keller, William Nix, Rowland Cannon, and Michael Drory. Some of this work was performed under a Cooperative Research and Development Agreement CRADA between Berkeley Lab and Hewlett Packard Laboratories. The Berkeley Lab portion of this work was supported by the Director, Ofce of Science, Ofce of Advanced Scientic Computing Research, Advanced Energy Projects and Technology Research AEPTR Division, of the U.S. Department of Energy under Contract No. DE-AC0376SF00098. We acknowledge support by DARPA Contract No. MDA972-96-3-0014.
1

S. Nakamura, The Blue Laser Diode: GaN Based Light Emitters and Lasers Springer, Berlin, 1997 . 2 J. W. Ager III, T. Suski, S. Ruvimov, J. Krueger, G. Conti, E. R. Weber, M. D. Bremser, R. F. Davis, and C. P. Kuo, Mater. Res. Soc. Symp. Proc. 449, 775 1997 . 3 K. Hiramatsu, T. Detchprohm, and I. Akaksaki, Jpn. J. Appl. Phys., Part 1 32, 1528 1993 .

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

7752
4 5

J. Appl. Phys., Vol. 87, No. 11, 1 June 2000

23

Romano et al. A. Atkinson, T. Johnson, A. H. Harker, and S. C. Jain, Thin Solid Films 274, 106 1996 . 24 S. C. Jain, H. E. Maes, K. Pinardi, and I. De Wolf, J. Appl. Phys. 79, 8145 1996 and references therein. 25 A. Polian, M. Grimsditch, and I. Grzegory, J. Appl. Phys. 79, 3343 1996 . 26 I. De Wolf, Semicond. Sci. Technol. 11, 139 1996 and references therein. 27 G. S. Cargill III, J. Angilello, and K. L. Kavanagh, Phys. Rev. Lett. 61, 1748 1988 . 28 J. Neugebauer and C. G. Van de Walle, Appl. Phys. Lett. 69, 503 1996 . 29 J. Neugebauer and C. G. Van de Walle, in Festkorperprobleme/Advances in Solid State Physics, edited by R. Helbig Vieweg, Braunschweig Germany, 1996 , Vol. 35, p. 25. 30 C. G. Van de Walle and J. Neugebauer, Appl. Phys. Lett. 70, 2577 1997 . 31 C. Kiesielowski et al., Phys. Rev. B 54, 17 745 1996 . 32 J. Kruger et al., Mater. Res. Soc. Symp. Proc. 482, 447 1998 . 33 M. F. Doerner and W. D. Nix, CRC Crit. Rev. Solid State Mater. Sci. 14, 225 1988 . 34 H. Windischmann, CRC Crit. Rev. Solid State Mater. Sci. 17, 547 1992 . 35 S. Tanaka, S. Iwai, and Y. Aoyagi, Appl. Phys. Lett. 69, 4096 1996 . 36 S. Keller, U. K. Mishra, S. P. DenBaars, and W. Siefert, Jpn. J. Appl. Phys., Part 2 37, L431 1998 . 37 A. C. Abare, M. Hansen, J. S. Speck, S. P. DenBaars, and L. A. Coldren, Electron. Lett. 35, 1559 1999 . 38 Amber Abare private communication .

C. Kisielowski et al., Phys. Rev. B 54, 17 745 1996 . S. Ruvimov et al., Appl. Phys. Lett. 69, 990 1996 . 6 I. Lee, I. Choi, C. Lee, E. Shin, D. Kim, S. K. Noh, S. Son, K. Lim, and H. J. Lee, J. Appl. Phys. 83, 5787 1998 . 7 S. Nakamura, T. Mukai, and M. Senoh, Jpn. J. Appl. Phys., Part 1 31, 2883 1992 . 8 D. P. Bour et al., IEEE J. Sel. Top. Quantum Electron. 4, 498 1998 . 9 S. Hearne, E. Chason, J. Han, J. A. Floro, J. Figiel, J. Hunter, H. Amano, and I. S. T. Tsong, Appl. Phys. Lett. 73, 356 1999 . 10 W. D. Nix and B. M. Clemens, J. Mater. Res. 14, 3467 1999 . 11 W. Gotz, R. S. Kern, C. H. Chen, H. Liu, D. A. Steigerwald, R. M. Fletcher, Mater. Sci. Eng., B 59, 211 1999 . 12 M. Leszczynski et al., Appl. Phys. Lett. 69, 73 1996 . 13 S. Porowski, J. Cryst. Growth 189190, 153 1998 . 14 J. W. Ager III, G. Conti, L. T. Romano, and C. Kisielowski, Mater. Res. Soc. Symp. Proc. 482, 769 1998 . 15 A. F. Wright, J. Appl. Phys. 82, 2833 1997 . 16 G. Stoney, Proc. R. Soc. London, Ser. A 82, 172 1909 . 17 W. E. Tefft, J. Res. Natl. Bur. Stand., Sect. A 70, 277 1966 . 18 J. W. Edington, Interpretation of Transmission Electron Micrographs Universities Press, 1975 . 19 L. T. Romano, B. S. Krusor, M. D. McCluskey, D. P. Bour, and K. Nuaka, Appl. Phys. Lett. 73, 1757 1998 . 20 L. T. Romano, M. D. McCluskey, C. G. Van de Walle, J. E. Northrup, D. P. Bour, M. Kneissl, T. Suski, and J. Jun, Appl. Phys. Lett. 73, 1999 . 21 J. E. Northrup and J. Neugebauer, Phys. Rev. B 53, R10 477 1996 . 22 S. M. Hu, J. Appl. Phys. 50, 4661 1979 .

Downloaded 11 Mar 2013 to 195.220.191.37. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions