- Email: [email protected]
Atomic layer epitaxy for the growth of heterostructure devices
Jout~,,~l of Crystal Growth 93 (1988) 195-200 North-Holland, Amsterdam
195
ATOMIC LAYER EPITAXY FOR THE GROWTH OF HETEROSTRUCTURE DEVICES
S.P. DENBAARS, P.D. DAPKUS, C.A. BEYLER *, A. HARIZ and K.M. DZURKO * * Department of Electrical Engineering and Center for Photonic Technology, University of Southern California, Los Angeles, California 90089.0483, USA
Atomic layer ep~taxy (ALE) is a regime of metalorganic vapor phase epitaxial growth in which uniform growth of ultra-thin epitaxial layers by a self-limiting monolayer by monolayer deposition process is achieved. In this paper ALE has been applied to the growth of single crystal GaAs, AlAs and GaAs/AIGaAs heterostructures and devices. Data are presented that show one monolayer uniformity in ultra-thin layers grown by ALE. The dependence of growth rate on reactant flows and temperature are described. Cleaved comer TEM analysis of ALE epitaxial layer thicknesses demonstrates the "digital" nature of the deposition process. The low temperature photolu.minescence (PL) of ALE grown GaAs quantum wells exhibit narrow line intrinsic luminescence with linewidths comparable to the best reported values by conventional MOCVD. Quantum well lasers with ALE grown active regions have been demonstrated with laser thresholds as low as 400 A/era 2.
1. Introduction
Atomic layer epitaxy (ALE) is a relatively new crystal growth technique for compound semiconductors which offers control of the growth process at the atomic level [1,2]. In contrast to conventional vapor phase growth, where the reactive gaseous precursors arrive simultaneously at the substrate, ALE is a stepwise deposition process in which alternate monolayers of each constituent dement in the compound are deposited using a self-limiting surface reaction mechanism. By employing a new regime of metalorganic chemical vapor deposition (MOCVD) growth, in which saturated surface reactions control the growth, it is possible to alternately deposit monolayers of column III and column V elements so that only one monolayer of the III-V compound semiconductor is deposited in every cycle of the deposition, ~3y utilizing this approach it is possible to produce extremely uniform films by repetition of this alternate exposure cycle. Therefore ALE can be considered a "digital" deposition process where the epitaxial layer thicknesses are determined * AT & T Bell Laboratories Scholar. ** Natural Sciences and Engineering Research Council of Canada Scl:olar.
primarily by the number of exposure cycles rather than being highly sensitive to such parameters as growth time and reactant flux. The technique was originally demonstrated for the growth of II-VI compounds and has achieved excellent results as a manufacturing process for ZnS electroluminescent displays [3]. ALE has recently been applied to the growth of III-V compounds [4-6]. The kinetics of the ALE growth process and the influence of ALE growth parameters on materials and device quality are now emerging. In this paper, we present data on the ALE growth kinetics of GaAs and AlAs utilizing trimethylgalllium, trimethylaluminum and arsine. The experimental kinetics are consistent with a simple adsorption model. In addition, we demonstrate that extremely uniform quantum wells of GaAs can be grown by ALE. Also, we have incorporated ALE GaAs into quantum well injection lasers that operate at room temperature by utilizing a hybridization of conventional MOCVD and ALE. 2. Experimental ALE of GaAs, AlAs, and A1GaAs is achieved by using conventional MOCVD precursors in a fe~r step gas injection sequence at low growth
0022-6248/88/$03.50 © Elsevier Science Publishers B.V. (North-Holland Physics Publishing Division)
196
S.P. De~Baarset aL / ALE for growth of heterostructure devices
temperatures. This sequence consists of a 1 s pulse of arsine, a 1 s pure hydrogen purge, a 1 s pulse of TMGa or TMAI, and a final ~ s purge of hydrogen. Since the average gas residence time is only 0.5 s at the high gas flow velocities employed in this work, minimal gas phase mixing of the precursors will occur. This has been confirmed by the observation that longer hydrogen purge times did not change the deposited film thickness. The gas source system utilized for these studies consisted of an atmospheric pressure vent/run MOCVD system which employs a pressure transducer and hydrogen makeup line to eliminate gas concentration fluctuations. The growth reactor is a small volume rectangular tube which is water cooled at the top of promote laminar gas flow. By using an infrared heating assembly rapid changes in the growth temperature were possible. The use of variable growth temperature in a run allowed us to grow the best quality GaAs/Al0.sGa0.sAs heterostructure by a hybridization of ALE and conventional MOCVD. The highest quality AIGaAs is grown by conventional MOCVD at 750°C, the temperature is then lowered to the surface controlled growth regime (445-500°C) where the quantum well active region is grown by ALE. Epitaxial layer thickness measurements of the GaAs quantum wells are made by transmission electron microscopy (TEM) analysis of the freshly cleaved single crystal comer and photoluminescence emission energy from the individual wells. Utilizing the natural cleavage planes (110) in GaAs, it is possible to obtain an extremely sharp comer, the thinnest portion of which can be penetrated by the electrons. Thus by orienting the edge of the sample at 45 ° with respect to the electron beam, a thin area exists near the corner allowing TEM analysis of layers in the cross section. This is relatively new technique [7], which eliminates the time consuming sample preparation usually required for TEM evaluation of semiconductor layer thickness measurements. Evaluation by comer TEM and photoluminescence results confirms that tile ALE growth process is controlled at the atomic level. Analysis of ALE AlAs films was performed by cleaved comer TEM, SEM and angle lapping measurements of thicker layers.
The effect of several growth variables on the ALE process was studied by growing quantum wells under various ALE conditions and separating the wells by 1 #m AlGa.As barriers so that only photoluminescence (PL) emission from the uppermost quantum well is observed. The sample was then etched with a calibrated etch to within 1000 A, of the next quantum well. Utilizing this measurement technique five single quantum wells (SQW) of varying growth parameters could be analyzed on the same sample produced in a single run, Layer thicknesses were estimated from peak emission wavelength expected from a KronigPenney model of QW emission energies. Accuracy of the layer thickness determined in this way was confirmed by cleaved comer TEM for thin layers, and by SEM analysis of thicker ALE GaAs layers.
3. Results and discussion
3.1. Experiment results The "digital" nature of the ALE growth process is best illustrate by the dependence of deposited film thickness on the number of deposition cycles as shown in fig. 1 where the solid line indica,es perfect monolayer deposition per cycle. Layer thickness measurements for this illustration were performed by cleaved corner TEM which is accurate to within 10% at the magnification used. In comparison with molecular beam epitaxy (MBE) and MOCVD, ALE can be considered a "digital process" since it does not require the
120 100
r, oWW,dt,(A)I
°g. ao 6O 40
20 0
0
10 20 30 Number of Exposure Cycles
40
Fig. 1. Film thickness dependence on number of ALE deposition cycles demonstrates "digital" nature of ALE growth process.
S.P. DenBaars et al. / A r -7,for growth of heterostructure devices
r °FT
O
0
Conv.MOCVD(T=650C,
[-o-
T=445°C
I;
T=:565:c
~
10
20 30 40 TMGa Inlected(id)
50
60
Fig. 2. Comparison of growth rate of GaAs grown utilizing ALE growth regime and conventional MOCVD.
precise control of such analog parameters as growth time and reactant flux. Saturated monolayer growth of GaAs has been achieved under conditions where surface reactions dominate the deposition of gallium and arsenic absorbates. In contrast to conventional MOCVD, in which the growth rate is directly proportional to the TMGa flux in the diffusion controlled re#me, the ALE growth rate is strongly sublinear in its dependence on both TMGa and arsine as shown in figs. 2 and 3. The conventional MOCVD data shown in fig. 2 has been normalized to the same delivery rate as the ALE data to illustrate this point. It is evident that at low temperatures, gas phase reaction rates decrease and surface kinetics control the growth rate. Note that the growth rate increases linearly with the reactant volume injected during the cycle up to a critical value near one monolayer and then saturates. The chemisorption of a Ga methyl species is proposed to explain this saturation with TM(3a exposure 4¸ A
2
MonoiayerGaAs(100)
=
Z: 1 I-
0
I
0
100
I
t
200 300 AmineInjected(~tl)
,
400
Fig. 3. Growth rate of ALE GaAs layers as a function of arsine injected into reactor.
197
over the temperature range 445-485 ° C. The fact that the deposited film thicknesses slightly exceed the ideal one monolayer per cycle deposition rate is thought to be caused by the deposition of involatile atomic gallium or incomplete flushing of the reactor. The strong saturation with arsine exposure shown in fig. 3 indicates that the arsine is not limiting the growth rate and the Ga deposition step is responsible for the nonideal saturation of fig. 2. In a previous study on the decomposition of TMGa in H2, data was presented which indicated that TMGa decomposes in two stages [8]. In the first stage, which occurred in the temperature range 380-450°C in the decomposition experiments, TMGa sequentially loses two methyl radicals in rapid succession. The third methyl radical is lost only in the second stage at higher temperatures to release atomic Ga. Thus it is likely that the partial gas phase decomposition of TMGa in H 2 at the temperatures and times employed here for ALE will result in the formation of CH3Ga. The saturated monolayer growth we have ob~e~ed suggests that the adsorbed species must retain some of the organic radicals to remain volatile enough to allow only one monolayer to chemisorb and avoid the deposition of excess Ga. This is confirmed by the fact that other researchers observe the formation of Ga droplets at higher temperatures [9]. Therefore, we conclude that CH3Ga is the surface species in our experiments. The quality of quantum wells formed by the ALE growth of the GaAs and MOCVD growth of Al0.sGao.sAs at higher temperatures is demonstrated in the 8 K photoluminescence spectra as shown in fig. 4. Note that all five quantum wells exhibit intrinsic radiative emission with narrow linewidths. The PL linewidths for each corresponding well thickness compare quite favorably with quantum wells grown in our laboratory or reported in the literature [10] by conventional MOCVD. The 1-3 meV increase in the linewidths of the ALE grown quantum wells in comparison to conventional grown MOCVD QWs is presumably caused by a small amount of impurity incorporation that may occur during the 3 min cool down. At 8 K, interface fluctuations and compositional fluctuations in the A1GaAs barrier are
S.P. DenBaars et al. / A L E for growth of heterostructure devices
198 I
I
8K PHOTOLUMINESCENCE
I
90~,
rneV Z
30~ 40~8~ I 6500
6900
7300 7700 WAVELENGTH (A)
)
8100
8500
Fig. 4. Photoluminescence spectra (8 K) of single quantum wells with ALE GaAs active regions.
known to be the predominant PL broadening mechanisms [11,12]. By utilizing a PL broadening model b.y Singh et al. [11] with a lateral island size of 100 A, a well width fluctuation of one monolayer can account for the 12.2 meV linewidth of the 30 ,~ quantum well grown by ALE. If compositional fluctuations are the dominant broadening mechanisms, the PL linewidths can be accounted for by only a 1.0% compositional fluctuation in the Al0.,~Ga0.sAs barrier layers [12]. Therefore, the PL spectra suggest that the interface abruptness and quality of the quantum wells has not been significantly altered by interrupting the growth to lower the temperature to ALE conditions. Excellent layer thickness uniformity is one of the inherent advantages expected from using the self-limiting monolayer gr,-,wth mechanism in the ALE process. We have realized excellent QW thickness uniformities through the use of the ALE deposition process. A variation of less than one monolayer in a 70 i, QW has been observed over an entire sample grown in our small scale research reactor [13]. The QW width variation across the wafer was measured by focusing the excitation laser to a 500 ~m diameter spot and measuring the peak emission energy shift as function of spot position. A quantum well grown by conventional MOCVD on the same untilted susceptor exhibited a 20-30% variation in layer thickness due to the gas depletion effect of TMGa along the gas flow direction.
Heterostructure devices that rely on highly uniform ultra-thin layers and abrupt interfaces would greatly benefit from this inherent advantage of ALE. For example, in the quantum well injection laser the lasing wavelength and threshold current density are strongly dependent upon the active region thickness in the 10-50 ,~, range. As a test vehicle for utilizing ALE in device structures, we have grown GaAs/A1GaAs quantum well lasers in which the active region is grown by ALE. Laser emission is observed from broad area devices operated under pulsed condition of 0.3 ms current pulses at a repetition rate of 10 kHz at room temperature. Lasers with a 100 ,~, quantum well deposited by ALE and undoped 2000 ,~ graded index layers have yielded the best results to date. As shown in table 1, threshold current densities of ALE grown separate confinement heterostructure lasers compare favorably with the best QW injection lasers grown in our lab. We anticipate even further reduction in threshold current densities upon further optimization of the confinement region and the ALE deposition process. Deposition of AlAs by ALE also demonstrates a monolayer saturation growth regime as shown by the dependence of growth rate on reactant flux as shown in fig. 5. The fact that AlAs growth saturates at slightly higher TMA1 quantities at 460°C than TMGa at 4 6 0 ° C is most probably due to the different decomposition rate constant of TMA1. This indicates that the partial gas phase decomposition of TMA1 also plays an important role in achieving ALE with metalorganic precursors. This is supported by the actuality that TMA1 is a dimer in the gas phase and one would expect less TMAI is needed for saturated growth. ALE of AlGa_As was also achieved by pulsed introduction of the gaseous precursors. Room temperature photoluminescence from both thick
Table 1 Threshold current densities Structure
SCH GRIN-SCH
Current density ( A / c m 2) Conventional MOCVD
ALE
400 160
640 400
S.P. DenBaars et al. / A L E for growth of heterostructure devices
199
4
(kGa) which predicts the amount of involatile Ga produced by the pyrolysis of monomethylgallium.
/
¢1
~2
f
dg
.u_ ¢. i- 1
}
d e / d t = k ~ d ~ ( 1 - e ) - k d ~ s ( e ) + k c , a,
MonolayerAIAs(IO0)
(1)
where i
2
.
I
4
.
i
.
kG,
=
6 8 TMAI Injected(p.I)
10
Fig. 5. Growth rate of AlAs layers grown under ALE conditions as a function of TMAI injected into the reactor.
A1GaAs ALE layers and quantum wells in which the A1GaAs barrier is also grown by ALE has been studied. These data show the expected behavior; however, low temperature luminescence of ALE AlGa.AS exh~bl.ts extrinsic luminescence indicating that further optimization of the deposition process is needed.
=
k MMGaCMMGa~//Ns ;
(2)
kGa utilizes the activation energy ( - 7 7 . 5 kcal/mol) required for the removal of the third methyl radical from T M G a which was determined by Jacko and Price [14]. The concentration of monomethylgallium (CMMG,) can be estimated from an earlier kinetic study in which removal of the first methyl from T M G a was observed to be the rate limiting step. CMMGa = Ng[1 - exp(--kTMGat)].
(3)
This leads to a Langmuir adsorption isotherm for the surface coverage. kads+kG" {1-exp[-(kaas+kdes)t]},
3.2. ALE growth model The ALE growth data observed by the use of MOCVD precursors can be qualitatively explained by a first order adsorption model of the surface controlled growth. We describe the rote of surface coverage ( d O / d r ) as the difference between a surface adsorption rate of a gallium methyl complex, kad,, and a surface desorption rate, kd~. To account for the observed deviation from perfect monolayer growth we introduce a rate expression
where both kad s and kde s a r e functions of temperature. When the species is physisorbed on the surface, the temperature dependence of (9 is controlled by kde s and d(9/dt decreases with increasing temperature. The data of figs. 2 and 6 show the opposite trend. This indicates that the adsorbed species are chemisorbed and that k a a s controls the temperature dependence of (9. We expect, therefore, that kad s will be dependent upon some gas phase or surface reaction that is more strongly activated than the surface desorption process. We
A
e
.t
O
I"]'=" ~' SurfaceKinetics ....._~./ L ControlledRegime
if 400
2 475°C
/14" 656,2TMG I ~ o
(4)
= kads + kde s
"1
I GaAs(100) I I 500 Temperature
.~'-
DiffusionC0ntr$11ed Growth Regirne I I
600
(°C)
Fig. 6. Effect of temperature on growth rate in surface reaction controlled ALE growth regime.
F-
0 Iv
0
~
465°C
I
,
1
20 40 TMGa Injected(id)
,
60
Fig. 7. ALE growth rate as predicted by first order adsorption model.
200
S.P. DenBaars et aL / ALE for growth of heterostructure devices
technologies. This has been demonstrated by the growth of high performance quantum well laser devices with threshold current densities as low as 400 A / c m 2. The kinetics for the ALE process have been determined and are shown to behave in accordance with a first order adsorption model.
Table 2 Rate constants k(TMGa) = 1 × 10 TM e x p ( - 60 kcal/RT) k(MMGa) = 5 x 1022 e x p ( - 77.5 k c a l / R T ) k (des) = 333 exp( - 4.6 kcal/RT)
assume that the diffusive flux of the absorbate is proportional to the reactant concentration in the gas phase, Ng (cm-3), proportional to the gas phase or surface reaction rate that produces the adsorbate, kTMGa (S-~-), and proportional to the diffusion distance which we take to be the hydrodynamic boundary layer, /J (cm). The effective adsorption is given by kads =
NgkTMGaS/Ns,
4. Conclusions In conclusion, saturated monolayer growth of GaAs and AlAs has been observed by using metalorganic precursors at growth temperatures where kinetics of surface reactions dominate the deposition process. We have confirmed some of the inherent advantages of the ALE technique by demonstrating the "digital" nature of the deposition process and growing extremely uniform high quality quantum wells. The approach we have Illi|i7Oq.
"l'h~ q ~ n 3 P ,
Tile authors gratefully acknowledge the program support of the Office of Naval Research and the Solar Energy Research Inst:,tute. 1he expertise of Jack Worrall in the TEM evaluation of the heterostructure samples is greatly appreciated.
(5)
where Ns is the surface site density for GaAs (100) in cm -2. Utilizing the rate constants shown in table 2 which have been determined from previous results published in the literature [8,14.15], we can model the ALE deposition process. As shown ir. the calculated curves of fig. 7, we have been able to qualitatively explain the observed ALE growth behavior.
tnkon
Acknowledgements
rlrOpllr~,r~
Ith~lt
ar~
r'nm_
monly employed in the M O C V D technology, allowing us to build on an existing technology base and to integrate the two approaches to permit hybrid structures that are grown partially by both
References [1] M. Pessa, R. Makela and R. Suntola, Appl. Phya. Letters 38 (1980) 131. [2] M. Pessa, P. Huttunene and M.A. Herman, J. Appl. Phys. 54 (1983) 604"L [3] C . H i . Goodman and M.V. Pessa, J. Appi. Phys. 80 (1986) R65. [4] J. Nishizawa, H. Abe and T. Kurabayashi, J. Electrochem. Soc. 132 (1985) 1197. [5] M.A. Tischler and S.M. Bedair, Appl. Phys. Letters 48 (1986) 1961. [6] M. Yoshida, H. Watanabe and F. Uesugi, J. Electrochem. Soc. 137 (1985) 677. [7] H. Kakibayashi and F. Nagata, 3urface Sci. 74 (1986) 84. [8] S.P. DenBaar~, B.Y. Maa, P.D. Dapkus, A.D. Danner and H.C. Lee, J. Crystal Growth 77 (1986) 188. [9] J. Nishizawa, T. Kurabayashi, H. Abe and A. Nozoe, Surface Sci. 185 (1987) 249. [10] D.C. Bertolet, J.K. Hsu and K.M. Lau, J. Appl. Phys. 62 (1987) 120. [11] J. Singh, K.K. Bajaj and S. Caudhuri, Appl. Phys. Letters 44 (1984) '~35. [12] S.B. Obale, A. Madhukar, F. Voillot, M. Thomsen, W.C. Tang, T.C. Lee, Y. Kim and P. Chen, Phys. Rev. B36 0987) 1662. [13] .q P DenBaar, C A Rey!er. A Hafz and P.D. Dapkus, Appl. Phys. Letters 51 (1987) 1530. [14] M.G. Jacko and S.J.W. Price, Can. J. Chem. 41 (1963) 1560. [15] Y. Aoyagi, A. Doi, S. Iwai. S. Namba, J. Vacuum Sci. Technol. B5 (1987) 1460.
195
ATOMIC LAYER EPITAXY FOR THE GROWTH OF HETEROSTRUCTURE DEVICES
S.P. DENBAARS, P.D. DAPKUS, C.A. BEYLER *, A. HARIZ and K.M. DZURKO * * Department of Electrical Engineering and Center for Photonic Technology, University of Southern California, Los Angeles, California 90089.0483, USA
Atomic layer ep~taxy (ALE) is a regime of metalorganic vapor phase epitaxial growth in which uniform growth of ultra-thin epitaxial layers by a self-limiting monolayer by monolayer deposition process is achieved. In this paper ALE has been applied to the growth of single crystal GaAs, AlAs and GaAs/AIGaAs heterostructures and devices. Data are presented that show one monolayer uniformity in ultra-thin layers grown by ALE. The dependence of growth rate on reactant flows and temperature are described. Cleaved comer TEM analysis of ALE epitaxial layer thicknesses demonstrates the "digital" nature of the deposition process. The low temperature photolu.minescence (PL) of ALE grown GaAs quantum wells exhibit narrow line intrinsic luminescence with linewidths comparable to the best reported values by conventional MOCVD. Quantum well lasers with ALE grown active regions have been demonstrated with laser thresholds as low as 400 A/era 2.
1. Introduction
Atomic layer epitaxy (ALE) is a relatively new crystal growth technique for compound semiconductors which offers control of the growth process at the atomic level [1,2]. In contrast to conventional vapor phase growth, where the reactive gaseous precursors arrive simultaneously at the substrate, ALE is a stepwise deposition process in which alternate monolayers of each constituent dement in the compound are deposited using a self-limiting surface reaction mechanism. By employing a new regime of metalorganic chemical vapor deposition (MOCVD) growth, in which saturated surface reactions control the growth, it is possible to alternately deposit monolayers of column III and column V elements so that only one monolayer of the III-V compound semiconductor is deposited in every cycle of the deposition, ~3y utilizing this approach it is possible to produce extremely uniform films by repetition of this alternate exposure cycle. Therefore ALE can be considered a "digital" deposition process where the epitaxial layer thicknesses are determined * AT & T Bell Laboratories Scholar. ** Natural Sciences and Engineering Research Council of Canada Scl:olar.
primarily by the number of exposure cycles rather than being highly sensitive to such parameters as growth time and reactant flux. The technique was originally demonstrated for the growth of II-VI compounds and has achieved excellent results as a manufacturing process for ZnS electroluminescent displays [3]. ALE has recently been applied to the growth of III-V compounds [4-6]. The kinetics of the ALE growth process and the influence of ALE growth parameters on materials and device quality are now emerging. In this paper, we present data on the ALE growth kinetics of GaAs and AlAs utilizing trimethylgalllium, trimethylaluminum and arsine. The experimental kinetics are consistent with a simple adsorption model. In addition, we demonstrate that extremely uniform quantum wells of GaAs can be grown by ALE. Also, we have incorporated ALE GaAs into quantum well injection lasers that operate at room temperature by utilizing a hybridization of conventional MOCVD and ALE. 2. Experimental ALE of GaAs, AlAs, and A1GaAs is achieved by using conventional MOCVD precursors in a fe~r step gas injection sequence at low growth
0022-6248/88/$03.50 © Elsevier Science Publishers B.V. (North-Holland Physics Publishing Division)
196
S.P. De~Baarset aL / ALE for growth of heterostructure devices
temperatures. This sequence consists of a 1 s pulse of arsine, a 1 s pure hydrogen purge, a 1 s pulse of TMGa or TMAI, and a final ~ s purge of hydrogen. Since the average gas residence time is only 0.5 s at the high gas flow velocities employed in this work, minimal gas phase mixing of the precursors will occur. This has been confirmed by the observation that longer hydrogen purge times did not change the deposited film thickness. The gas source system utilized for these studies consisted of an atmospheric pressure vent/run MOCVD system which employs a pressure transducer and hydrogen makeup line to eliminate gas concentration fluctuations. The growth reactor is a small volume rectangular tube which is water cooled at the top of promote laminar gas flow. By using an infrared heating assembly rapid changes in the growth temperature were possible. The use of variable growth temperature in a run allowed us to grow the best quality GaAs/Al0.sGa0.sAs heterostructure by a hybridization of ALE and conventional MOCVD. The highest quality AIGaAs is grown by conventional MOCVD at 750°C, the temperature is then lowered to the surface controlled growth regime (445-500°C) where the quantum well active region is grown by ALE. Epitaxial layer thickness measurements of the GaAs quantum wells are made by transmission electron microscopy (TEM) analysis of the freshly cleaved single crystal comer and photoluminescence emission energy from the individual wells. Utilizing the natural cleavage planes (110) in GaAs, it is possible to obtain an extremely sharp comer, the thinnest portion of which can be penetrated by the electrons. Thus by orienting the edge of the sample at 45 ° with respect to the electron beam, a thin area exists near the corner allowing TEM analysis of layers in the cross section. This is relatively new technique [7], which eliminates the time consuming sample preparation usually required for TEM evaluation of semiconductor layer thickness measurements. Evaluation by comer TEM and photoluminescence results confirms that tile ALE growth process is controlled at the atomic level. Analysis of ALE AlAs films was performed by cleaved comer TEM, SEM and angle lapping measurements of thicker layers.
The effect of several growth variables on the ALE process was studied by growing quantum wells under various ALE conditions and separating the wells by 1 #m AlGa.As barriers so that only photoluminescence (PL) emission from the uppermost quantum well is observed. The sample was then etched with a calibrated etch to within 1000 A, of the next quantum well. Utilizing this measurement technique five single quantum wells (SQW) of varying growth parameters could be analyzed on the same sample produced in a single run, Layer thicknesses were estimated from peak emission wavelength expected from a KronigPenney model of QW emission energies. Accuracy of the layer thickness determined in this way was confirmed by cleaved comer TEM for thin layers, and by SEM analysis of thicker ALE GaAs layers.
3. Results and discussion
3.1. Experiment results The "digital" nature of the ALE growth process is best illustrate by the dependence of deposited film thickness on the number of deposition cycles as shown in fig. 1 where the solid line indica,es perfect monolayer deposition per cycle. Layer thickness measurements for this illustration were performed by cleaved corner TEM which is accurate to within 10% at the magnification used. In comparison with molecular beam epitaxy (MBE) and MOCVD, ALE can be considered a "digital process" since it does not require the
120 100
r, oWW,dt,(A)I
°g. ao 6O 40
20 0
0
10 20 30 Number of Exposure Cycles
40
Fig. 1. Film thickness dependence on number of ALE deposition cycles demonstrates "digital" nature of ALE growth process.
S.P. DenBaars et al. / A r -7,for growth of heterostructure devices
r °FT
O
0
Conv.MOCVD(T=650C,
[-o-
T=445°C
I;
T=:565:c
~
10
20 30 40 TMGa Inlected(id)
50
60
Fig. 2. Comparison of growth rate of GaAs grown utilizing ALE growth regime and conventional MOCVD.
precise control of such analog parameters as growth time and reactant flux. Saturated monolayer growth of GaAs has been achieved under conditions where surface reactions dominate the deposition of gallium and arsenic absorbates. In contrast to conventional MOCVD, in which the growth rate is directly proportional to the TMGa flux in the diffusion controlled re#me, the ALE growth rate is strongly sublinear in its dependence on both TMGa and arsine as shown in figs. 2 and 3. The conventional MOCVD data shown in fig. 2 has been normalized to the same delivery rate as the ALE data to illustrate this point. It is evident that at low temperatures, gas phase reaction rates decrease and surface kinetics control the growth rate. Note that the growth rate increases linearly with the reactant volume injected during the cycle up to a critical value near one monolayer and then saturates. The chemisorption of a Ga methyl species is proposed to explain this saturation with TM(3a exposure 4¸ A
2
MonoiayerGaAs(100)
=
Z: 1 I-
0
I
0
100
I
t
200 300 AmineInjected(~tl)
,
400
Fig. 3. Growth rate of ALE GaAs layers as a function of arsine injected into reactor.
197
over the temperature range 445-485 ° C. The fact that the deposited film thicknesses slightly exceed the ideal one monolayer per cycle deposition rate is thought to be caused by the deposition of involatile atomic gallium or incomplete flushing of the reactor. The strong saturation with arsine exposure shown in fig. 3 indicates that the arsine is not limiting the growth rate and the Ga deposition step is responsible for the nonideal saturation of fig. 2. In a previous study on the decomposition of TMGa in H2, data was presented which indicated that TMGa decomposes in two stages [8]. In the first stage, which occurred in the temperature range 380-450°C in the decomposition experiments, TMGa sequentially loses two methyl radicals in rapid succession. The third methyl radical is lost only in the second stage at higher temperatures to release atomic Ga. Thus it is likely that the partial gas phase decomposition of TMGa in H 2 at the temperatures and times employed here for ALE will result in the formation of CH3Ga. The saturated monolayer growth we have ob~e~ed suggests that the adsorbed species must retain some of the organic radicals to remain volatile enough to allow only one monolayer to chemisorb and avoid the deposition of excess Ga. This is confirmed by the fact that other researchers observe the formation of Ga droplets at higher temperatures [9]. Therefore, we conclude that CH3Ga is the surface species in our experiments. The quality of quantum wells formed by the ALE growth of the GaAs and MOCVD growth of Al0.sGao.sAs at higher temperatures is demonstrated in the 8 K photoluminescence spectra as shown in fig. 4. Note that all five quantum wells exhibit intrinsic radiative emission with narrow linewidths. The PL linewidths for each corresponding well thickness compare quite favorably with quantum wells grown in our laboratory or reported in the literature [10] by conventional MOCVD. The 1-3 meV increase in the linewidths of the ALE grown quantum wells in comparison to conventional grown MOCVD QWs is presumably caused by a small amount of impurity incorporation that may occur during the 3 min cool down. At 8 K, interface fluctuations and compositional fluctuations in the A1GaAs barrier are
S.P. DenBaars et al. / A L E for growth of heterostructure devices
198 I
I
8K PHOTOLUMINESCENCE
I
90~,
rneV Z
30~ 40~8~ I 6500
6900
7300 7700 WAVELENGTH (A)
)
8100
8500
Fig. 4. Photoluminescence spectra (8 K) of single quantum wells with ALE GaAs active regions.
known to be the predominant PL broadening mechanisms [11,12]. By utilizing a PL broadening model b.y Singh et al. [11] with a lateral island size of 100 A, a well width fluctuation of one monolayer can account for the 12.2 meV linewidth of the 30 ,~ quantum well grown by ALE. If compositional fluctuations are the dominant broadening mechanisms, the PL linewidths can be accounted for by only a 1.0% compositional fluctuation in the Al0.,~Ga0.sAs barrier layers [12]. Therefore, the PL spectra suggest that the interface abruptness and quality of the quantum wells has not been significantly altered by interrupting the growth to lower the temperature to ALE conditions. Excellent layer thickness uniformity is one of the inherent advantages expected from using the self-limiting monolayer gr,-,wth mechanism in the ALE process. We have realized excellent QW thickness uniformities through the use of the ALE deposition process. A variation of less than one monolayer in a 70 i, QW has been observed over an entire sample grown in our small scale research reactor [13]. The QW width variation across the wafer was measured by focusing the excitation laser to a 500 ~m diameter spot and measuring the peak emission energy shift as function of spot position. A quantum well grown by conventional MOCVD on the same untilted susceptor exhibited a 20-30% variation in layer thickness due to the gas depletion effect of TMGa along the gas flow direction.
Heterostructure devices that rely on highly uniform ultra-thin layers and abrupt interfaces would greatly benefit from this inherent advantage of ALE. For example, in the quantum well injection laser the lasing wavelength and threshold current density are strongly dependent upon the active region thickness in the 10-50 ,~, range. As a test vehicle for utilizing ALE in device structures, we have grown GaAs/A1GaAs quantum well lasers in which the active region is grown by ALE. Laser emission is observed from broad area devices operated under pulsed condition of 0.3 ms current pulses at a repetition rate of 10 kHz at room temperature. Lasers with a 100 ,~, quantum well deposited by ALE and undoped 2000 ,~ graded index layers have yielded the best results to date. As shown in table 1, threshold current densities of ALE grown separate confinement heterostructure lasers compare favorably with the best QW injection lasers grown in our lab. We anticipate even further reduction in threshold current densities upon further optimization of the confinement region and the ALE deposition process. Deposition of AlAs by ALE also demonstrates a monolayer saturation growth regime as shown by the dependence of growth rate on reactant flux as shown in fig. 5. The fact that AlAs growth saturates at slightly higher TMA1 quantities at 460°C than TMGa at 4 6 0 ° C is most probably due to the different decomposition rate constant of TMA1. This indicates that the partial gas phase decomposition of TMA1 also plays an important role in achieving ALE with metalorganic precursors. This is supported by the actuality that TMA1 is a dimer in the gas phase and one would expect less TMAI is needed for saturated growth. ALE of AlGa_As was also achieved by pulsed introduction of the gaseous precursors. Room temperature photoluminescence from both thick
Table 1 Threshold current densities Structure
SCH GRIN-SCH
Current density ( A / c m 2) Conventional MOCVD
ALE
400 160
640 400
S.P. DenBaars et al. / A L E for growth of heterostructure devices
199
4
(kGa) which predicts the amount of involatile Ga produced by the pyrolysis of monomethylgallium.
/
¢1
~2
f
dg
.u_ ¢. i- 1
}
d e / d t = k ~ d ~ ( 1 - e ) - k d ~ s ( e ) + k c , a,
MonolayerAIAs(IO0)
(1)
where i
2
.
I
4
.
i
.
kG,
=
6 8 TMAI Injected(p.I)
10
Fig. 5. Growth rate of AlAs layers grown under ALE conditions as a function of TMAI injected into the reactor.
A1GaAs ALE layers and quantum wells in which the A1GaAs barrier is also grown by ALE has been studied. These data show the expected behavior; however, low temperature luminescence of ALE AlGa.AS exh~bl.ts extrinsic luminescence indicating that further optimization of the deposition process is needed.
=
k MMGaCMMGa~//Ns ;
(2)
kGa utilizes the activation energy ( - 7 7 . 5 kcal/mol) required for the removal of the third methyl radical from T M G a which was determined by Jacko and Price [14]. The concentration of monomethylgallium (CMMG,) can be estimated from an earlier kinetic study in which removal of the first methyl from T M G a was observed to be the rate limiting step. CMMGa = Ng[1 - exp(--kTMGat)].
(3)
This leads to a Langmuir adsorption isotherm for the surface coverage. kads+kG" {1-exp[-(kaas+kdes)t]},
3.2. ALE growth model The ALE growth data observed by the use of MOCVD precursors can be qualitatively explained by a first order adsorption model of the surface controlled growth. We describe the rote of surface coverage ( d O / d r ) as the difference between a surface adsorption rate of a gallium methyl complex, kad,, and a surface desorption rate, kd~. To account for the observed deviation from perfect monolayer growth we introduce a rate expression
where both kad s and kde s a r e functions of temperature. When the species is physisorbed on the surface, the temperature dependence of (9 is controlled by kde s and d(9/dt decreases with increasing temperature. The data of figs. 2 and 6 show the opposite trend. This indicates that the adsorbed species are chemisorbed and that k a a s controls the temperature dependence of (9. We expect, therefore, that kad s will be dependent upon some gas phase or surface reaction that is more strongly activated than the surface desorption process. We
A
e
.t
O
I"]'=" ~' SurfaceKinetics ....._~./ L ControlledRegime
if 400
2 475°C
/14" 656,2TMG I ~ o
(4)
= kads + kde s
"1
I GaAs(100) I I 500 Temperature
.~'-
DiffusionC0ntr$11ed Growth Regirne I I
600
(°C)
Fig. 6. Effect of temperature on growth rate in surface reaction controlled ALE growth regime.
F-
0 Iv
0
~
465°C
I
,
1
20 40 TMGa Injected(id)
,
60
Fig. 7. ALE growth rate as predicted by first order adsorption model.
200
S.P. DenBaars et aL / ALE for growth of heterostructure devices
technologies. This has been demonstrated by the growth of high performance quantum well laser devices with threshold current densities as low as 400 A / c m 2. The kinetics for the ALE process have been determined and are shown to behave in accordance with a first order adsorption model.
Table 2 Rate constants k(TMGa) = 1 × 10 TM e x p ( - 60 kcal/RT) k(MMGa) = 5 x 1022 e x p ( - 77.5 k c a l / R T ) k (des) = 333 exp( - 4.6 kcal/RT)
assume that the diffusive flux of the absorbate is proportional to the reactant concentration in the gas phase, Ng (cm-3), proportional to the gas phase or surface reaction rate that produces the adsorbate, kTMGa (S-~-), and proportional to the diffusion distance which we take to be the hydrodynamic boundary layer, /J (cm). The effective adsorption is given by kads =
NgkTMGaS/Ns,
4. Conclusions In conclusion, saturated monolayer growth of GaAs and AlAs has been observed by using metalorganic precursors at growth temperatures where kinetics of surface reactions dominate the deposition process. We have confirmed some of the inherent advantages of the ALE technique by demonstrating the "digital" nature of the deposition process and growing extremely uniform high quality quantum wells. The approach we have Illi|i7Oq.
"l'h~ q ~ n 3 P ,
Tile authors gratefully acknowledge the program support of the Office of Naval Research and the Solar Energy Research Inst:,tute. 1he expertise of Jack Worrall in the TEM evaluation of the heterostructure samples is greatly appreciated.
(5)
where Ns is the surface site density for GaAs (100) in cm -2. Utilizing the rate constants shown in table 2 which have been determined from previous results published in the literature [8,14.15], we can model the ALE deposition process. As shown ir. the calculated curves of fig. 7, we have been able to qualitatively explain the observed ALE growth behavior.
tnkon
Acknowledgements
rlrOpllr~,r~
Ith~lt
ar~
r'nm_
monly employed in the M O C V D technology, allowing us to build on an existing technology base and to integrate the two approaches to permit hybrid structures that are grown partially by both
References [1] M. Pessa, R. Makela and R. Suntola, Appl. Phya. Letters 38 (1980) 131. [2] M. Pessa, P. Huttunene and M.A. Herman, J. Appl. Phys. 54 (1983) 604"L [3] C . H i . Goodman and M.V. Pessa, J. Appi. Phys. 80 (1986) R65. [4] J. Nishizawa, H. Abe and T. Kurabayashi, J. Electrochem. Soc. 132 (1985) 1197. [5] M.A. Tischler and S.M. Bedair, Appl. Phys. Letters 48 (1986) 1961. [6] M. Yoshida, H. Watanabe and F. Uesugi, J. Electrochem. Soc. 137 (1985) 677. [7] H. Kakibayashi and F. Nagata, 3urface Sci. 74 (1986) 84. [8] S.P. DenBaar~, B.Y. Maa, P.D. Dapkus, A.D. Danner and H.C. Lee, J. Crystal Growth 77 (1986) 188. [9] J. Nishizawa, T. Kurabayashi, H. Abe and A. Nozoe, Surface Sci. 185 (1987) 249. [10] D.C. Bertolet, J.K. Hsu and K.M. Lau, J. Appl. Phys. 62 (1987) 120. [11] J. Singh, K.K. Bajaj and S. Caudhuri, Appl. Phys. Letters 44 (1984) '~35. [12] S.B. Obale, A. Madhukar, F. Voillot, M. Thomsen, W.C. Tang, T.C. Lee, Y. Kim and P. Chen, Phys. Rev. B36 0987) 1662. [13] .q P DenBaar, C A Rey!er. A Hafz and P.D. Dapkus, Appl. Phys. Letters 51 (1987) 1530. [14] M.G. Jacko and S.J.W. Price, Can. J. Chem. 41 (1963) 1560. [15] Y. Aoyagi, A. Doi, S. Iwai. S. Namba, J. Vacuum Sci. Technol. B5 (1987) 1460.