GaAs/AlₓGa₁₋ₓAs Quantum Well Lasers Grown on GaAs and Si by Molecular Beam Epitaxy - CaltechTHESIS

measurements now are converging onto the 60:40 rule. It has been widely agreed
that the valence band discontinuity is

-31-

0.15

0.30

0.44

0.48

4'1

1.02±0.01

1.13±0.02

1.23±0.02

1.27±0.02

z

1.02±0.01

1.12±0.02

1.22±0.02

1.24±0.02

0.13±0.01

0.24±0.02

0.34±0.02

0.38±0.02

0.13±0.01

0.23±0.02

0.33±0.02

0.35±0.02

0.69±0.05

0.64±0.05

0.62±0.05

0.63±0.05

0.69±0.05

0.61±0.05

0.60±0.05

0.58±0.05

6Ec
6Ec
6E"
~Eg

~EC b
~Eg

a: 6Ec = , - o

b; 6Ec = 4>2 -o

Table 2.3 The conduction band discontinuity fl.Ee obtained from Schottky barrier heights for
different aluminum mole-fractions.

fl.Ev [Al As/ G aAs] =0 .50e V

2. Ge/ZnSe, ZnSe/GaAs, and GaAs/Ge. These three pairs provide a test of transitivity, i.e.,

substituting in the experimental data obtained by different people, we actually get
tlEv[Ge ~ ZnSe] + tlEv[ZnSe-+ GaAs] + tlEv[GaAs-+ Ge] = 0

-32-

0.31-------+----+-----"7""'-t-----1

eo
0.2t------+----....,-t------,-.-------i
0 .-=------'------'------+------__.
0. I
0.3
0.4

Au

GaAs

(b)

Figure 2.4 ( a) The conduction band offset .6..Ec can be obtained by <1>1 - <1> 0, and (b) the two
energy bands can be properly aligned by shrinking the thickness of gold in between.

3. InAs/GaSb system.

This is a very unusual line-up, the conduction band edge

of InAs is above 150meV below the valence band edge of GaSb, an example of
broken-gap line-up. The available experimental data seem to indicate that
.6..Ev[InAs/GaSb] = 0.5leV.

The above five heterosystems together can verify the validity of a band line-up
theory since they include group IV, III-V, and II-VI elements all together. The newly

-33-

proposed theory presented earlier, performs very well under these tests. Notice that
GaAs/AlAs is the most reliable system today and the only one whose numerical band
offset has been experimentally determined beyond any doubt. The experiment value of
0.50eV is very close to my calculated value of 0.54eV. Even the recent Tersoff theory
does not meet this test: it predicts a value of 0.35eV. As in Table 2.2, the proposed
theory also works very well with other experimental data collected so far.
The only exception is for InAs/GaSb system. A possible reason is the following:
First, InAs/GaSb has a lattice constant mismatch of 8%. Whenever there is a large
lattice mismatch, the line of reasoning we have followed so far, which assumes a perfect termination of a infinite lattice, will fail. Green's function calculation can not be
performed since we do not know the potential near the interface caused by lattice mismatch, and we do not know how to expand the inverse Hamiltonian (Green's operator).
Second, it maybe similar to the history of the GaAs/ AlAs system, which was first believed to have almost a zero valence band offset (85:15 rule) that was changed later,
and therefore the current experimental data may not be reliable. More careful measurements will clarify this issue and ascertain the value of experimental band line-up
of InAs/GaSb system.

§ 2.7 Conclusions
In this chapter, we have given Tersoff's quantum dipole theory a theoretical basis
using Green's function technique, since the wavefunction matching can be done by a
secular equation of Green's functions on the interface. It is further noticed that the
secular equation when spatially averaged, gives rise to the Tersoff model previously
based only on some qualitative physical arguments. Furthermore, we have imposed
a k-selection rule on the Tersoff theory, and successfully applied it to the r point (k
=0) to eliminate the uncertainty of Terso-ff's model due to the dependence on R. This

-34-

simplified theory works better than the Tersofftheory when compared with experiments,
especially; it predicts the band offset of AlAs/GaAs system correctly. A photoelectric
measurement is made on both the Schottky barrier height of Au/ AlGaAs and the
conduction band offset !:!,,Ee between GaAs and AlGaAs as a function of aluminum
mole fraction x.

-35-

§ 2 .8 References

[l] T. Lukes, Solid State Theory, ed. P. T. Landsberg, John Wiley & Sons, London,
1969, part F, and references therein.
[2] F. Flores, Nuovo Cimento, 14B, 11(1973).
[3] F. Garcia-Moliner and J. Rubio, J. Phys. C (Solid State), 2, 1789(1969).
[4] J. Tersoff, Phys. Rev. , B30, 4874(1984) ,Phys. Rev. B32, 6968 (1985).
[5] W. A. Harrison, J. Vac. Sci. Technol., B3(4), 1231(1985).
[6] E. 0. Kane, J. Phys. Chem. Solids, 1, 83(1956).
[7] W. A. Harrison, J. Vac. Sci. Technol. , 14, 1016(1977).
[8] H. Z. Chen, H. Wang, A. Ghaffari, H. Marko<;, and A. Yariv, Appl. Phys. Lett. ,
51, 990(1987), and references therein.

[9] H. Kroemer, VLSI Electronics: Microsiructure Science, 10, 121(1985).
[10] R. Dingle, Advances in Solid-State Physics, H. J. Queisser, ed. , 15, 21 (1975).

[ll] H. Kroemer, J. Vac. Sci. Technol. , B2, 433 (1984).
[12] D. Arnold, A. Ketterson, T. Henderson, J. Klem, and H. Morkoc;, J. Appl. Phys. 57,
2880 (1985).
[13] T. W. Hickmott, P. M. Solomon, R. Fischer and H. Marko<;, J. Appl. Phys. 58,
2853 (1985).
[14] J. Batey, S. L. Wright, and D. J. DiMaria, J. Appl. Phys. 58, 484 (1985).
[15] W. I. Wang and F. Stern, J. Vac. Sci. Technol. B3, 1280 (1985).
[16] M. A. Haase, M. A. Emanuel, S. C. Smith, J. J. Coleman, and G. E. Stillman,
Appl. Phys. Lett. 50, 404 (1987).
[17] C. A. Mead, Solid-State Electon. 9, 1023 (1966).
[18] W. Monch, Surf. Sci. 132, 92 (1983).
[19] H. Kroemer, Conference Workbook of 2nd International Conference on Modulated
Semiconductor Structures, Sept. 9-13, Kyoto, Japan, 1985, Supplement p. 797.

-36-

!0] W. A. Harrison and J. Tersoff, J. Vac. Sci. Technol. B4, 1068 (1986).
\1] H. Hasegawa and H. Ohno, J. Vac. Sci. Technol. B4, 1130 (1986), Jpn. J. Appl. Phys. 25,
1265 (1986).
:2] G. Margaritondo, Surf. Sci. 168, 439 (1986).
:3] C.R. Crowell, W. G. Spitzer, L. E. Howarth, and E. E. LaBate, Phys. Rev. 127,
2006 (1962).
4] H. Unlu and H. Marko~, Bull. of American Physical Society 32, No. 3, (1987).
5] J. S. Best, Appl. Phys. Lett. 34, 522 (1979).
6] C. A. Mead and W. G. Spitzer, Phys. Rev. 134, A713 (1964).

-37-

Chapter 3

MEE Growth of GaAs-on-GaAs Quantu1n Well Lasers

§ 3.1 An introduction

In this chapter, techniques to optimize crystal growth in an MBE system and our
basic understanding of how a quantum well laser operates are combined as we try to
grow low-threshold quantum well lasers.

Details of MBE growth of GaAs/ AlGaAs

quantum well lasers on GaAs substrate (the work on Si is discussed in the next two
chapters) are presented. The main purpose of this chapter is to discuss various useful
techniques ( dirty tricks) used in MBE growth that were developed in this research and
eventually enabled us to grow the lowest threshold lasers in the world. Emphasis is
made on how to experimentally optimize growth conditions and minimize the effect of
minor instrument failures. Results obtained from quantum well lasers are also analyzed
to provide a simple experimental procedure for growing low-threshold lasers. Topics
such as the basic growth processes, mathematical models for the growth, etc., have
been thoroughly covered by many authors in many excellent articles [1], and will not
be discussed here.

-38-

§ 3.2 Substrate preparation

It is self-evident for any materials scientist that a clean substrate is the single
most important thing for any epitaxial growth. Contaminated substrates can lead to
mistakes that are regrettable. Not long ago, the most common macroscopic defects on
(100) GaAs layers grown by MBE, the "oval defects," were attributed to the effect of
Ga "spitting" from the Ga effusion cell. Condensed Ga at the orifice of the Ga crucible
can roll back, splash into the Ga melt, and cause eruption and ejection of Ga droplets,
and were believed to form "oval defects." Therefore, many users of MBE kept the Ga
crucible filled as close to the orifice as possible, or provided additional heating to Ga to
eliminate "oval defects." But their success was very limited. More careful investigations
by many groups have now revealed that the "oval defects" are not at all related to any
Ga "spitting." Instead, gallium oxide in the melt, and especially carbon on the GaAs
substrate surface are the true causes of this defect [2]. A good cleaning procedure would
have prevented all these unnecessary troubles.

Cleaning procedure
The procedure we have been using is the same one used by Morkoi; group at the
University of Illinois, and it is shown in Appendix II.

Substrate mounting
At this point, we have a clean GaAs substrate with perhaps a very thin layer of
some unknown com pound ( a few atomic layers, and the layer is so thin that a clear
RHEED pattern from the sample inside the MBE growth chamber can be observed
without heating the substrate). We should immediately transfer the substrate to a
clean fume hood where it can be In-mounted to a Mo block. Since the mounting is
done at 250°C, a thin layer of oxide is grown as a result of the reaction of GaAs with
oxygen in air (Figure 3.1). A common misconception is that this protective oxide layer
is grown in the last water rinsing. The substrate after the chemical cleaning can be

-39-

recontaminated by dirty air in the room. A protective thin oxide layer has to be grown
in the air inside a clean fume hood in a clean room. Furthermore, this layer blocks
RHEED pattern below 580°C and provides the only reliable calibration of substrate
temperature for GaAs substrate (oxide on GaAs desorbes at 580°C). A GaAs substrate
directly mounted (see Chapter 6 for Si substrate mounting) on a Mo block without
use of In, does not have this thermally grown oxide layer, and consequently a RHEED
pattern can be observed even at room temperature (the solution of course, is to put
the mounted substrate in a heated oven to grow this oxide layer). The In mounted
substrate is now safe, and it can wait for some time to be loaded into MBE, although
it should be done as quickly as possible.

§ 3.3 Growth of GaAs/ AIGaAs quantum well lasers

Most of the lasers used in this thesis research employ a graded refractive index separate confinement heterostructure (GRINSCH) single quantum well (SQW) structure.
The details of the growth are described in the following.

Pregrowth preparation of MBE
After the substrate is mounted on a Mo block and loaded into the MBE system, it
is first heated to 300° C in the loading chamber (first of the three chambers) to remove
any water vapor condensed on it. The pressure in the loading chamber may briefly
rise, but it will come down quickly. Then the substrate is transferred to the analysis
chamber where the vacuum is about two orders of magnitude higher than that in the
loading chamber. After waiting for some time until the pressure is as low as before, the
substrate is transferred to the growth chamber.
At this time, the growth chamber, which contains a liquid nitrogen cooled shroud
surrounding it, has reached a high vacuum of about 5 x 10- 9 torr or lower. V/hen the
base pressure in the growth chamber goes down to 10- 9 torr, the outgasing of all the

-40-

Heater: 250 't

Figure 3.1 Thermal growth of a thin protective oxide film in clean air.

source materials in effusion cells to be used in growth should be started. This is done
by heating the cells to a temperature slightly higher than the growth temperature to
blow off any condensations accumulated on the surface of source materials. The arsenic
cell is heated up very slowly because any excess heating can result in large amount of
arsenic loss, so it should be started with shroud cooling at the same time. \Vhen all the
materials have been outgased and reset to their growth temperatures (from here on we
use a computer program to control their temperatures), and the As cell has reached its
set temperature, we can start the growth procedure.

-41-

Thermal cleaning of GaAs substrate
The temperature of the GaAs substrate is controlled manually by a power supply.
At each increase of heating power, both current I and voltage V going into heating
filaments under the Mo block are recorded. The power P = IV is increased when the
thermal couple reading stops rising ( usually this takes a few minutes with In-mounted
substrate, and a few seconds with In-free mounting). When the heating power reaches
a certain level (this has to be based on previous growth records and is usually when
the substrate temperature is about 500° C), the RHEED instrument and pyrometer are
turned on to start monitoring the substrate surface. At about 580°C, the (2x4) and
C(2x8) surface reconstruction patterns appear clearly on the RHEED screen, which
indicates the desorption of the thin oxide layer. A brief temperature increase is used
to assure the complete desorption of oxide. The As shutter is opened at the same time
to prevent any loss of As atoms from the GaAs substrate. When the RHEED pattern
becomes sharp, an atomically clean surface is obtained and ready for growth.

RHEED pattern and surface reconstruction
A RHEED pattern appears when the substrate surface becomes clean. The identification of a RHEED pattern requires knowledge of the surface reconstructions. The
following notations are often used [3]:

1. GaAs (100) - (mxn) means that a GaAs crystal is orientated with the (100)
direction normal to the surface, and has a surface structure whose unit mesh
is mx n times larger than the underlying bulk unit cell.
2. If the mesh is centered, the notation would be GaAs (100) - C(mx n).

If

the mesh is rotated, the notation would be specified by the angle of rotation,
e.g., GaAs (111) - (-JI§x -JI§)R23.4°. If the surface reconstruction is an Asstabilized surface, we denote it by, e.g., GaAs (100) - C(4x2)As. GaAs (100)
- C(4x2)Ga is likewise used for an Ga-stabilized surface.

-42-

As an example, Figure 3.2 shows two of the most commonly observed surface
structures, (2x4) and C(2x8), in real and reciprocal space. Experimentally, it has been
known that [4] surface structures depend on three things: the background pressure,
whether the substrate is cooling down or heating up, and the temperature at which the
substrate is cooling down or heating up. For our laser growth, the substrate is heating
up, at 580°C, in a pressure of about lxl0- 7 Torr in an As-rich environment. According
to Cho [4], for a GaAs (100), (2x4)As(Ga) and C(2x8)As(Ga) reconstruction patterns
should be observed under normal growth conditions. It is very difficult to distinguish
these two patterns. The reason is illustrated in Figure 3.3 which shows reciprocal lattice
sections for both (2 x 4) and C( 2 x 8) and their expected RHEED patterns in different
azimuthal angles. For all practical purposes, one should rotate the substrate manually
and observe all the above patterns as a sign of a clean GaAs (100) surface, just after
the oxide protection layer is desorbed at 580°C.
As 2 pressure

It is known [1] that As 2 molecules ( not As 4 ) are responsible for the MBE growth
of GaAs. Arsenic molecules from a standard Knudsen cell are predominantly As 4 • A
subsequent "cracking" of As 4 which occurs when they arrive at GaAs surface, produces
As2 molecules which participate in GaAs growth. In practice, a PAs 4

= 10- 7 torr is

needed from a conventional As 4 cell that has no cracking effect at all and a PAs 2

= 10-

torr is needed from an arsenic cracker cell that has a high cracking-efficiency of about
90%. Arsenic pressure below this level will produce Ga-stabilized growth, which is
inconsistent with the As-terminated (100) substrate used.
GaAs buffer layer

When the oxide layer is desorbed, the substrate temperature is usually at 620-640°C
which is slightly higher than what is optimum. Therefore, we lower the temperature to
about 580-600°C and start the growth by opening the Ga shutter and desired dopants.

Real Space

Reciprocal Space

Iat !

(2 X 4)

•J

rrr
-x-rrrx7

(110)

(100)

-----x-x-x-----

(110)

,!C(ZLI
---..-0-0-0---0000
-lt-0-0-0----1

l I

Figure 3.2 The (2 x 4) and C(2 x 8) surface structures in real and reciprocal space.

Since we usually use n+ GaAs substrates, Ga and Si shutters are opened.

At the

opening of Ga shutter, a sudden drop of total pressure is observed on the ion gauge
indicating Ga atoms are combining with As atoms.
A few minutes into the growth the RHEED pattern should be checked again. A
good start of the growth usually shows sharp and streaky lines in RHEED patterns
which indicate a typical two-dimensional surface reconstruction.

O: Bulle

0: (2 X 4) and

C(2 X 8) both
I::,.: (2 X 4)
@: C(2 X 8)

(2 X4)

C(2 X 8)

I I
fl I I I
I I I I I I I I
11

(2X4)

/q2 XS)

'l

(')

00

(110)

Figure 3.3 The (2 x 4) and C(2 x 8) surface structures in reciprocal space and the associated

RHEED patterns in different azimuth in real space.

The GaAs buffer layer growth takes about an hour before the growth of AlGaAs is
started. Some believe that a thicker buffer layer will improve the quality of later AlGaAs
growth since AlGaAs will be farther away from the first interface where there might be
some defects and/or impurities. However, there has been no convincing evidence. The
best lasers that we have grown have a l.5µm GaAs buffer layer. Thicker buffer layers
have not resulted in any noticeable improvement in this study. This is because inside an

-45-

MBE's growth chamber, everything reaches thermal equilibrium quickly. Furthermore,
the residual impurities can not be removed by one more hour of growth anyway.
The As cell temperature should be kept low as long as the growth is still in an
As-stabilized environment (which can be observed on the RHEED screen).
Some researchers believe that the use of a superlattice region in the buffer layer
will improve the quality of growth. This is simply a misunderstanding of the dynamic
nature of an MEE growth. Although the use of a superlattice may temporarily trap
some impurities and stop defects from propagating for the time being, as soon as the
superlattice is finished, the growth returns to its previous condition quickly. The correct
moment to use a superlattice is just before the quantum well growth, to smooth the
interface and stop impurities and defects. Such is the case of our laser growth where
5 quantum wells have been used just before the active quantum well, to smooth the
interface and prepare it for the next quantum well growth [5].

AIGaAs cladding layer
Near the end of the GaAs buffer layer and when the growth approaches the beginning of the AlGaAs cladding layer, the As cell temperature is raised by 2 degrees to
raise the As pressure since the growth rate of AlGaAs is normally l.5µm/h, compared
to l.0µm/h for GaAs, and consumes more As. At the moment the Al shutter opens,
the substrate temperature is suddenly raised to 720°C by increasing the heating power.
From our experience over several hundred growths, P = llOTtV for low temperature (;:::::
600°C) and P = 225W for high temperature (;::::: 720°C) should be used. This is very
important since high substrate temperature can improve the optical quality of AlGaAs
dramatically [6], while the low temperature growth gives best crystal uniformity and
electrical properties. The two keys to a high quality AlGaAs growth are (a) high substrate temperature, and (b) minimum As to Ga ratio [6]. The substrate temperature
throughout the entire growth of a GRINSCH laser is shown in Figure 3.4.

-46-

Growth

Started

AI Ccell
Closed

Al Ccell
Opened

120°c

Oxide Off
at 580°C

600°C

time

ti)

1c.,

c.,

Cl)

c.,

ti)

c.,

ti)

c.,

Figure 3.4 The temperature profile for an entire laser growth.

Laser structure

After l.5µm of AlGaAs cladding layer growth, the growth of the GRIN region starts.
The refractive index can be changed by varying the Al mole fraction according to a prewritten computer program, to create a parabolic-like optical waveguide. The control
parameters of the Al cell temperature regulator have to be set so that it can respond
as quickly as possible to the temperature setting commands. The Si cell temperature is
reduced gradually for a doping level from l.Ox 10 18 cm- 3 in AlGaAs outside the GRIN,
down to l.Oxl0 17 cm- 3 when it is closed just before the quantum well. Three to five

-47-

5A wide smoothing quantum wells 1000 A

away from the active quantum well are

grown to improve surface quality. The Si shutter is closed about 100 A away from the
quantum well since Si atoms diffuse in AlGaAs. The Be shutter is opened after the
quantum well is grown, and its temperature is gradually raised for a doping level from
l.0x 10 17 cm- 3 up to l.0x 10 18 cm- 3 , while the Al mole fraction is increased to form the
second half of GRIN. Figure 3.5 shows the above structure schematically.
The upper l.5µm of AlGaAs cladding layer was grown after this, followed by a
p-GaAs capping layer doped by Be to 5.0xl0 19 cm- 3 • The substrate temperature of the
last GaAs layer growth is dropped abruptly to 600°C.

Al profile: a technological issue
Ever since the first GRINSCH laser, it has been assumed that a parabolic GRINSCH structure is fundamentally superior [7]. From a simple analysis, however, the
profile of the Al concentration does not appear important. A carefully designed variation in Al mole fraction creates an optical waveguide that greatly reduces light scattering losses. The exact shape of this index waveguide, though, is not important, nor
is it possible to obtain due to random fluctuations of Al effusion cell. As long as the
waveguide has a lateral dimension that is com parable to the size of fundamental optical
mode, its purpose is well served. Different waveguide shapes which are created either
intentionally or unintentionally by fluctuations of Al cell from time to time should give
different threshold current, yet no sign of such effect has ever been reported. What has
been consistently observed, however, is that the threshold current is always low when
an MBE system is clean and working well, regardless the shape of the waveguide; and
when the MBE is not working well, no structure can achieve a low-threshold current.
Therefore, there seems to be a discrepancy between our analysis and the experiment.
The answer, as is always the case, lies in the technological aspect of growth. It has
been well established experimentally that the growth of an "inverted" GaAs ( GaAs

-48-

0.4µm GaAs

1750,l{ Alx Ga 1_xAs

x=o.e-x=0.2

eoX GaAs
1750.8. Al x Ga 1_1 As

x=0.2 __,. 0.6, 5 Quantum wells

2.

Figure 3.5 Schematic drawing of the conduction band edge of a GRINSCH laser.

on AlGaAs) is often of poor quality [8). The higher the Al concentration in AlGaAs,
the poorer GaAs on it. Therefore, it is possible that a GRINSCH laser has a better
GaAs quantum well since the Al concentration is gradually reduced, giving the surface
enough time to reconstruct. Moreover, if one simply grows a step index waveguide
(Figure 3.6) where the Al mole fraction is changed abruptly, the quality of the "inverted" GaAs quantum well is not good. Clearly, the difference between a GRINSCH

-49-

Poor Interface
Quality

Figure 3.6 The band edge diagram of a step-indexed laser. The abrupt aluminum mole fraction
changes result in a poor MBE growth of the following layer.

and a step-index laser is mainly technological.

Once the technological difference is

eliminated, the two should produce similar results. This, has been recently confirmed
by a distinguished Russian research group at Leningrad Institute led by Dr. Garbusov
[9], totally unaware of our record threshold current density of 80 A/cm 2 • They have
succeeded using the very difficult LPE technique, to grow a step-index single quantum
well (100 A) laser and obtained a threshold current density of 98 A/cm 2 • Their result
confirms the above analysis very convincingly.

Technique of growth interruption
It has become increasingly popular in the MBE community to apply growth interruption techniques to improve interface quality. The purpose of a growth interruption is
to allow the ongoing interface to relax ( or to reconstruct) following a change in atomic
composition (such as Al grading in AlGaAs). Such a technique can improve the quality

-50-

of "inverted" GaAs on AlGaAs. The procedure for the (Ga,Al)As system is shown in
Appendix III.

This technique has been used in the GaAs-on-Si growth, and it improves the surface
morphology noticeably.

§ 3.4 Fabrication and measurement of broad area lasers

Fabrication
Each time a sample is grown, measurements are performed to obtain the threshold
current density, Jth, and the lasing wavelength A. A low-threshold current density is
very important since it is a universally accepted measure of MBE growth quality: the
lower the threshold, the better the crystal growth. The fabrication procedure of broad
area lasers is shown in Appendix IV.

Measurement of threshold current density
The measurement of threshold current of a broad area laser device is done with
a pulsed power supply since continuous wave (CW) operation is impossible without
mounting the laser upside down on a heat sink.

Typical current pulses are 100 ns

wide and are at a 50 kHz rate. The laser diode is forward biased in front of a Si p-i-n
photodiode. The pumping current signal ( converted to voltage by a current probe) and
p-i-n diode signal are fed into an oscilloscope. Light power vs. pumping current is
plotted for the laser until well above threshold. The intercept of the linear portion of
the L - I curve with current axis is the threshold current Ith, which is then divided by
the surface area ( measured by an optical microscope ) of the laser to obtain Jth. The
slope of the L - I curve gives the quantum efficiency: 77

= slope(W/A)/ Eg.

The measurement of lasing wavelength is done with an optical fiber bundle replacing
the p-i-n detector to collect light from the laser, which is then fed into a monochrometer
and collected by an optical signal multichannel analyzer which displays 90 - 100 A of

-51-

the entire spectrum of the optical signal. When the laser is biased just below threshold,
a large number of Fabry-Perot cavity modes are visible on the analyzer (Figure 3.7).
At threshold, one such cavity mode spikes up suddenly atop the spontaneous emission
background. The lasing wavelength and the threshold current can thus be determined
simultaneously. This threshold is then compared with the value obtained form L - I
measurement. The spectrum measurement often gives a threshold value slightly higher
than given by the L - I method and is believed to be more reliable.

§ 3.5 The effect of substrate misorientations

The previous discussions do not involve the type of substrate used in the MBE
growth. In practice, however, they do make a significant difference. The various crystal
orientations are shown in Figure 3.8.
The surface of a commercially available substrate is not atomically flat. It contains
microscopic steps that expose mini-surfaces along different orientations. For example,
on a primarily (100) oriented substrate, there exist (lll)Ga, (lll)As, (211)Ga, etc.,
mini-surfaces, although most surface area is covered by (100) oriented mini-surfaces.
Since atomic adsorption depends on the surface structure, some substrate orientations
are energetically favored during an MBE growth. On the other hand, the impurity
trapping rate is relatively independent of the substrate orientation. Therefore, a slight
substrate tilt can expose a considerable amount of atomic steps that will help the crystal
growth without increasing impurity trapping.
To study the effect of substrate tilting on laser performance, we have very often
used in this thesis work (100) substrates tilted toward the nearest (lll)Ga plane by
4°. The substrate tilting creates microscopic steps (Figure 3.9) on the surface. The
ratio of the height of the steps to the separation of steps determines the angle of tilting.
The growth rate of ( Al,Ga)As along these steps is believed higher than that on a

-52-

Osscilascope

Power supply
£',

v,xx,xx

xxx

I)
" '>' Y, ))

X )<

JC
lr'X~,rv,nnr
)( X

)( )( )( f

)(

)( X X

xxxx

~ y

J<

'J

+1.0

L{

{t

Cu
(a)
Spectrometer

(b)

Figure 3. 7 Schematic drawing of the experimental set-up used to measure ( a) threshold current
and (b) optical spectrum.

straight (100) surface and may be responsible for the improved surface morphology and
reduced impurity trapping. Furthermore, at these steps, open Ga bonds are exposed,
and as a result, arsenic molecules can make additional bonds to these extra Ga atoms,
thereby increasing the effective arsenic sticking coefficient. Consequently, the impurity
incorporation rate is lowered because of higher arsenic incorporation.
The effect of substrate misorientation has been used successfully to improve the

4°

➔(111} Ga \ 4°

Figure 3.8 Schematic drawing of the various crystal orientations of GaAs. At the center of the
drawing is a GaAs (100) substrate tilted 4° toward the Ga plane.

quality of "inverted" layer growth where GaAs is grown on AlGaAs in a standard HEMT
structure [10]. The tilting also improves photoluminescence (higher light intensity and
narrower linewidth) from a quantum well [11]. Therefore, we further investigated the
effect of substrate tilting on quantum well laser performance and the possibility of
replacing all untilted substrates we were using with the tilted ones.
4° tilted and straight (100) substrate are chemically cleaned in the same process.

-54-

o GALLIUM

(001)

L,. o,

(1

t 0)

-{i10]

[Ho). (•ARSENIC-TYPE

f Li

STEP EDGE•

L=moJ
{001)
(HI)

L,,;o,
(110)

Figure 3.9 Schematic drawing of the atomic steps caused by substrate tilting in ( a) tilted toward

the (111) Ga plane, and (b) tilted toward the (111) As plane.

-55-

Then they are In-mounted side by side on the same Mo block. The two substrates
receive exactly the same treatment throughout the entire growth.
The MEE growths have revealed very dramatic differences. First, when the growth
condition is optimum, e.g., the substrate temperature for AlGaAs growth is 720°C, the
two substrates have produced nearly identical surface morphologies which only slightly
favor the tilted one and they produce similar threshold current densities in broad area
lasers, with the tilted one slightly lower. When the growth condition is not optimum,
however, the tilted substrate always shows better surface morphology, even when the
surface of the straight one appears so bad that it would normally prompt us to abort
the growth. On several occasions, for example, when the substrate temperature was
purposely set at 660-680°C (in the "forbidden window" of AlGaAs growth between
640°C and 700°C), the straight substrate could not produce any lasing device, while
the tilted one produced lasers with moderately higher threshold current densities ( see
Table 3.1 ). Table 3.2 shows the experimental data on lasers obtained from straight and

tilted substrates at optimum growth temperature of 720°C.
Appropriate use of substrate tilting has proven effective in improving crystal quality and device performance. The use of substrate tilting, however, is not limited to
improving the quality of large area crystal growth alone; it can also be used to fabricate devices such as quantum wire lasers [12] using a modified version of MEE which is
called "enhanced mobility epitaxy" that deposits Ga and As atomic layers alternately.
Furthermore, under optimum conditions, it helped to grow the world's lowest threshold
current density lasers (lt1, of 80 A/cm 2 ). Now it has become a common practice to use
tilted substrates to improve MEE growth quality in many laboratories.

Effect of quantum well thickness on threshold current density
Quantum well thickness has been considered important to low-threshold current
density because of its effect on transition energy. Calculations on the optimum quan-

-56-

Temperature( CC,

JttfNcm2,

(100)

680

472

tilted

680

163

(100)

660

not lasing

tilted

660

403

Substrate

Table 3.1 Laser characteristics versus substrate temperature.

tum well thickness for laser performance are common in the literature. Many of our
early experimental results, however, suggest that the thickness of the quantum well
should have very little effect on threshold current density. To clarify this issue, lasers
with different quantum well thickness are grown under optimum conditions. Table 3.2
shows the results of threshold current densities versus quantum well thicknesses. The
threshold current density is almost constant in the range of 65-125 A. The small fluctuations are caused by the order of growth: 65A, 125A, 35A, and 95A (later growth
should be better since MBE is cleaner). This contradicts the previously accepted conclusion that at 70 A , a noticeable drop in threshold current density can be observed
[13]. We attribute their result to the relatively low quality of laser materials used in
their study. Higher threshold current densities must contain unknown and undesirable
effects which might be larger than the effect under investigation, and therefore can
not be used to check theory. The observed relationship between 1th and Lz becomes
apparent if a correct threshold condition is adopted.

-57-

Substrate

L (A)

L(mm)

J1b(Ncml

(100)

165

3.20

130

tilted

165

3.29

115

(100)

125

3.26

114

tilted

125

3.21

93

(100)

95

3.18

148

tilted

95

3.09

120

(100)

65

3.19

124

tilted

65

3.24

117

Table 3.2 Laser characteristics versus quantum well width at substrate temperature of 720° C for
both tilted and straight (100) substrates.

Physically, when the quantum well thickness is very small, for example, Lz :S 30A,
the ratio of well depth to well width increases, and so does the transition energy from
the electron ground state to the hole ground state E 1c - Eihh. (Table 3.3 shows the
energy levels of the electron, the heavy hole and light hole, the transition energy, and
the measured lasing wavelength of quantum well lasers with Lz from 35 A to 480 A. The
calculation uses a self-consistent model detailed in [8].) As a result, two things happen:
first, the transparency condition that requires the separation of Fermi levels l:,.ef; to be
larger than the transition energy nw, requires an increase in the separation between
the Fermi levels to balance the increase in transition energy; second, more unbounded
bulk states outside the quantum well are being filled as the Fermi levels move closer

-58-

mea

L z(A) E 1/meV) E hJmeV) E (meV) Egapllli

35

91

29

51

1536

1536

65

50

13

30

1479

1485

95

31

19

1454

1445

125

20

13

1440

1433

165

13

1432

1423

300

1421

1401

480

1419

1400

exciton binding energy is assumed to be 8meV

Table 3.3 The energy levels of the electron, the heavy hole and light hole, the transition energy,
and the measured lasing wavelength of quantum well lasers with Lz from 35 A to 480 A.

to conduction and valence band edges, thereby increasing the recombination current in
the GRIN region substantially.

If we denote the necessary increase in the Fermi level difference by 6¢, then the
excess recombination current density lexces, will have to be increased by a factor of
exp((J6rp) according to the standard p-n junction theory, where /J is a constant depending

on tern perature. Using the calculated energies of E 1 c and Eihh, it can easily be verified
that lexcess(35A_) is 40 times of lexcess(l25A_) and 100 times of lexcess(480A_). This is
why it does not pay to use a very thin quantum well as active layer. The gain from
reducing the quantum well volume is offset by the effective reduction of quantum well
depths t:i.Ec and t:i.Ev through exp((Jt:i.Ec) and exp((Jt:i.Ev)- In other words, there is a

-59-

trade-off between quantum well volume and carrier confinement. On the other hand,
an individual carrier can experience yet another effect that only concerns the behavior
of a single electron known as the real space electron transfer effect in which case the
electron wavefunctions in a quantum well will tail into the higher bandgap AlGaAs
region, thereby reducing the number of carriers in the well. However, this effect is only
important when Lz < 30A. 30Ais a borderline also because the fluctuation in Lz is about
two atomic layers, ~ 10 A.
On the other hand, when Lz

---+

oo, two things can also happen to increase the

threshold: first, the condition under which two-dimensional bimolecular recombination occurs in a two-dimensional plane is changed, and carriers now recombine in a
three-dimensional space with a higher density of states; second, the nature of electron
wavefunctions begin to change so that the differential gain coefficient g 2 D ( usually larger
than g 3 D) has to be replaced by g 3 D.
The transition from 2D to 3D is a gradual process. As far as device performance is
concerned, it occurs when Jth begins to increase linearly with Lz, which has been observed (Figure 3.10) experimentally. Physically, the transition from 2D to 3D represents
a process in which carriers in the ground state of a 2D system begin to occupy the second and higher quantized energy levels. To estimate the transition quantum well width,
we can define (this is somewhat subjective) the quantum well width L;rans to be one for
which the total number of electrons in ground state (i=l) equals to the total number
of electrons in second and higher (i=2,3 ...... ) quantized states, n 1

= n 2 + n 3 + ...... , and

solve it numerically. The result, 160A, is very close to the experimental turning point
that is shown in Figure 3.10.

A more rigorous treatment
The well width dependence of gain and threshold current in a quantum well laser
can be deduced from a treatment given by Yariv [14] on the gain in a quantum well

-60-

900
G)
(2)

800

@)

700

---

1:

L=3170,3244,3090,3300,3297,3140,301 Oµ
L=1190, 1144, 1060, 1085, 1068, 1240,990µ
L-575,530,590,520,568,500,420 µ
L-330,300,260,300,366,316,31 o µ

600

(.)

500

.r;

-,

400

300
200
100

35

65 100 125

165 200

300

400

480 500

Lz (A)
Figure 3.10 Experimental data showing the relationship between threshold current density 1th
and quantum well width Lz. The four curves represent four different cavity lengths, 0.3mm, 0.5mm,
1.0mm, and 3.0mm.

laser which is based on a rigorous density matrix formalism [15). This treatment also
includes the effect of gain broadening not discussed here.
We start with the expression for the complex susceptibility of an electronic material
well below optical gain saturation [14)
(3.1)

x(w)
and the gain is given by
1 (w)

= - ~x"(w)
n2

(3.2)

The calculation involves the counting of N 2 -N1 in (3.1). Since a single electron state

-61-

in a crystal is characterized by its wave vector k, the counting of N2 - N 1 is translated
into an integration over k's satisfying certain conditions. The first condition is the
k-selection rule. In a perfect crystal, the electronic momentum Tik is much larger than
the photon momentum 1ikphoton; therefore, in the independent electron approximation
[16], kc

= kv holds for an optical transition. In a crystal containing a large number of

lattice mismatch or impurity defects, the k-selection rule is likely to be violated since
the impurities can take up an appreciable amount of electronic momentum and as a
result, kc -:ft kv. This situation has been analyzed by Landsberg et al. [17], and used
by Saint-Cricq et al. [18], to calculate the gain and the threshold current in quantum
well lasers.
In a real laser, the situation is somewhat in between the two limiting cases. In the
following, we will try to apply both rules in the gain calculations and then com pare
them to experimental data. The difference is in the counting of the number of electrons
that will make a transition from the conduction band to the valence band with k value
between kc and kc + dkc in conduction band, and between kv and kv + dkv in valence
band.
In the case of k-selection rule,

(3.3)

but in the case of non-k-selection rule, each electron in the conduction band has the
additional possibility of recombining with any hole in the valence band. And there are
a total of

vx

JPv1:)

fv(E)dE

(3.4)

holes for each electron in the conduction band. Furthermore, the non k-selection rule
results in a modification of the transition probability [19,40] by a factor of µenv. So the

-62-

combined effect gives for the non-k-selection case,

d(N2 - N1lt:k'

= j dEv V

L L
Pc(Ec)

Pv(Ev)
(3.5)

X [fc(Ec)- fv(Ev)] lµenvl dEc
The exponential gain coefficient is thus

(3.6)

for k-selection, and

(3.7)

The well width dependence is simply

,(w/=k'
,(wl#'

Lz
L2z

(3.8)

where

Ev

hwo

Pr,c,v

1 + e(E-n)/ kT
Em ao
-6471" X (-e-)3 X (1 + a6k 2 )V
me

ao

0.52810- 8 cm

H(x)

l if x > 0, H(x)

0 if X < 0

-63-

and hw = photon energy, µ = dipole matrix element, T 2 = intraband relaxation time,
and nr = refractive index. All the quantities are in CGS units.
The difference between the k-selection and non-k-selection rules on the gain can be
estimated as follows
Pr

(3.9)

taking
fl.E

h/r

me

0.067 me

mv

0.48 me

lµenv 1

~ 64 71" X ( cmeao )3
me

we get
(me+ mv)64(cmeao/mc) 3

(3.10)

~ 30 [...!:..:_] [...!._]
100.A

lns

A rigorous numerical calculation yields a ratio of 6.02 instead of 30.
The gain calculated using non-k-selection rule, therefore, is significantly lower than
using k-selection. This agrees with the general prediction made by other authors [17,18]
that the advantage of a quantum well laser is only apparent when k-selection is assumed.

Transparency sheet carrier density
To reach lasing, the quantum well material has to become optically transparent,
that is, ,(w)=O. According to (3.2), this happens when fe(Ec) - fv(Ec - hw 0 )=0. In two
dimensions, the density of states for i-th subband is given by

Pi

(3.11)

where mi is the effective mass of the carrier in i-th subband. The total carrier density

-64-

is thus

= ~ Pi fc(Eci)dEci
= kTL Pi ln[l + eC4>n-Ec,)/kT]

(3.12)

for electrons, and a similar expression for holes.
Since the energy separation between the first (i= 1) and the second (i=2) levels is
larger than 3kT for a typical quantum well (Lz :S 150.A) used in GRINSCH lasers, and
the main contribution to the Fermi functions fc and fv are from the ground states, only
the ground states (i= 1) will be considered. Therefore,
(3.13)

and similarly
(3.14)

Since n = p, the transparency condition fc - fv=0 becomes
(3.15)

Notice that the equation does not depend on Lz and this is why the transparency current
density for a ideal quantum well laser is independent of the well thickness. Using the
following values for GaAs, me= 0.067me, mv = 0.48me, and T = 300K, the sheet carrier
density n can be determined from (3.15)
n(mc, mv, T)

(3.16)

Recombination lifetime
Both theory [20,21] and experiment [22,23] indicate that the recombination rate
as well as lifetime are constant; furthermore, Christen et al. [22], have measured the
lifetime r: r = 7.0 ns for an undoped 110 A quantum well, and r = 3.25 ns for a

-65-

p-doped 1.0x 10 16 cm- 3 quantum well. Matsusue et al. [24] have found the lifetime in
undoped quantum wells to be 5 ns. The background doping in our MBE system was
measured to be 1.0x 10 16 cm- 3 of p type. Therefore, we take r = 3 ns as given by
Christen [22].
Thus the transparency current density for an ideal 2D GaAs is

= ne/r

Jo

and r = 3 ns, J 0

(3.17)

= 63A/cm- 2 •

This value of J 0

= 63A/cm- 2 is in good agreement with the transparency current

density calculated by Thom phoson [25] for bulk GaAs using a strict k-selection rule
(Thomphoson obtained J 0 = 4000A/cm 2 µm-1, which would give for 100 A J 0 = 40A/cm 2 ).
This illustrates the major advantage of quantum well lasers: a large reduction of the
transparency current. And this advantage is not offset by the confinement factor r QW
discussed below since the optical field is very weak below the transparency.

Above transparency
So far it is shown that a J 0 which only depends on the material parameters, is
required to make the quantum well transparent. Above the transparency, optical field
becomes strong in the cavity, and the modal gain experienced by an optical mode
propagating along the junction plane is modified by a confinement factor
(3.18)

where the confinement factor I'Qw is given by

QW

JL,/2 IEl2dz
_-_L-•_/_2_ __
f~oo IEl2dz

(3.19)

The modal gain -y~=;,t is therefore independent of well width Lz; however, the modal
kfk'
gam -Ymode depends on Lz
f-tk'(
'i'mode W)

Lz

~ 2Lz X - - ex Lz .
Wmode

(3.20)

-66-

To obtain the threshold current density, the modal gain -Ymode(w) is multiplied by a
factor g that is independent of Lz [24]. The factor g can be easily measured from 1th
versus Lz curve for quantum well lasers of different width. Our measurements yielded
a single value of g

= 0.7 A/cm which is roughly independent of Lz.

The total threshold current density for both k-selection and non k-selection cases
are plotted in Figure 3.11 and compared to experiment. It is clear that the k-selection
rule fits experimental data and should be observed. This calculation differs from the
one by Saint-Cricq [18] which does not use the correct matrix element that includes the
effect of envelope wavefunction. It is also in qualitative agreement with the calculations
assuming a k-selection rule [26,27,28,29,30,31,32]. The most striking feature of the nonk-selection rule is a linear increase of the threshold current density with quantum well
width. This, however, has not been observed in our measurements. And since all the
lasers used in this study have a undoped quantum well, it is safe to conclude that the
non-k-selection rule does not hold for undoped materials.

Higher order corrections due to band-mixing
The previous calculation can be regarded as a first order approximate theory which
is valid in the two-dimensional and perfect crystal limit. A rigorous theory, however,
has to take into account the effect of band-mixing as a result of realistic band structures

[33]. Recently, more basic studies [34,35,36,37] have shown that with more detailed
analysis, the band structure and optical matrix elements of the quantum well can be
quite different from the ones used in the previous calculations (equations (3 .6) ,(3. 7)).
The dispersion relationship E(k) can be numerically computed from a more fundamental

k · p method [32,36,37], and the density of states can be obtained by
A(E) =

dS

S;(E) 47r 3 JdE(k)/dkl

where S;(E) 1s the surface m k space obeying the equation E;(E)

(3.21)

E, for the i-th

-67-

1200

1000

800

600

g=30cm-1

400

~~

200

o-50

~ I

k- selection

100

200

300

Figure 3.11 The calculated threshold current density versus quantum well width for both kselection and non-k-selection cases.

subband. The surface integral in the limit of Lz -+ 0, is

p;(E)

41r 3 jdE(k)/dkl

_k_,

dk_ I
1rLz dE;(k)

(3.22)

This density of states can be substituted into equations (3.3), (3.5), (3.6) and (3.7) to
calculate gain.

-68-

1501------+-------------------------1

'E

.......

100

"'

.r:.
-,....

50

10

15

20

25

1/L log 1/R (cm-1 )

Figure 3.12 The plot of threshold current density versus inverse cavity length for one of the best

GRINSCH lasers obtained in this research. The threshold current density is well below 100 A/cm2 .

Experimental determination of Jo
If the cavity length is long enough, the mirror loss will be small, and the threshold
current density will approach the transparency current density. During this study, the
lowest threshold current density ever obtained in any semiconductor laser has been
demonstrated. Figure 3.12 shows the threshold current density versus reciprocal cavity
length for the best laser device obtained in this study:
1th

= 98A/ cm 2

L = 520µm

SOA/cnl

L = 3300µm

(3.23)

and an external quantum efficiency of 85%.

-69-

The portion of current density due to mirror loss and free carrier absorption can
be estimated

The gain coefficient g was measured to be 0.7 A/cm, a=15/cm, and R=3.6.
L=3.3mm, J=20A/cm 2 •

For

The measured transparency current density is 60 A/cm 2 ,

agrees well with the calculated value of 63 A/cm 2 •
Another indication that our quantum well material is of very high quality, is demonstrated by the fact that when mounted upside down on a diamond heat sink, one such
laser can put out nearly 3 Watts of CW power from a 100 µm stripe, and it can also
be biased to a record high of 130 times above its threshold.

Other substrate orientations
Recently, Hayakawa et al.[38], have used (111 )As 0.5° --. (100) substrates to grow
quantum well lasers and obtained threshold current density as low as 120 A/cm 2 • They
attributed the good result to heavier hole effective mass along (lll)As and concluded
that the differential gain can be improved. This argument presents some problems,
since the ideal situation for a high differential gain coefficient is to have the conduction
band and valence band with similar effective masses. This is explained in the following:

where Ne,v

(3.25)

and

m~(v are effective density of states. We require that first, r/Jn - r/Jp = nw,

and second, lr/Jn - Eel

+ lr/Jp - Ev I be minimized. Furthermore, p ~ n at threshold. The

second requirement will give a minimum current density.

Using equation (3.1) and

treating me and mv as parameters, it is easy to show that:

me mv
me

+ 111v

= Const.
mznzmum

(3.26)

(3.27)

-70-

will happen only when me= mv.
Physically, when a laser is forward biased, the separation of electron and hole quasiFermi levels ¢n and c/Jp reaches the value of transition energy as soon as ¢n and c/Jp move
into the conduction and valence band, respectively. A very large hole effective mass
slows down the movement of c/Jp, so that ¢n has to move significantly into the conduction
band to satisfy transparency condition c/Jn - c/Jp

= nw. The electron Fermi level located

deep in the conduction band can give rise to a large threshold current, which is exactly
what we don't want. Therefore, the success of (111) substrate must be due to the
improved MBE growth of AlGaAs crystal on (111) oriented facets [39] in general.
The possibility of many other tilting orientations is under investigation currently.

Keys to a low-threshold laser growth
Here we summarize the key factors to a low-threshold laser growth (it also applies
to any other device).

First, the quantum well width Lz should be small enough so that two dimensional effects dominate, but large enough so that the interface quality is not affected.
A working range from 30 to 150 A is available depending on the choice of lasing
wavelength.
Second, the substrate temperature TsT should be kept at 600° C for GaAs growth
and 720°C for AlGaAs growth. An indium-free direct heating mount can be used to
further improve the required rapid change in substrate temperature.

Third, there are now at least two other types of substrate that produce lower
threshold current density results than the conventional (100) oriented substrate:

(i)

(100) 4° _, (lll)Ga

(ii)

(lll)As 0.5° _, (100)

-71-

§ 3.6 Conclusions

In this chapter, we have described the details of GaAs/ AlGaAs quantum well laser
growth, the fabrication and measurement of broad area lasers, and the effect of substrate
misorientation. It is shown that a proper substrate tilting can greatly improve the
MBE growth of GaAs and AlGaAs under all circumstances, especially when the growth
condition is not optimum.

The use of tilted substrates has resulted in the lowest

threshold current density ever reported for any semiconductor lasers. The experimental
value of 80 A/cm 2 for a 3.3mm long laser is an improvement of a factor of 3 over the
previous record.
The effect of quantum well width on threshold current density is analyzed using
both the k-selection rule and the non-k-selection rule. It is found that in an undoped
material, the k-selection rule gives a very good fit of the experimental data. Furthermore, k-selection rule gives a modal gain that is independent of Lz, compared to the
non-k-selection result that shows a L-; 1 dependence.
The transparency current density Jo is calculated using k-selection rule, which compares very favorably with experimental results (5% difference).

-72-

§ 3.7 References

[1] C. T. Faxon, B. A. Joyce, and M. T. Norris, J. Cryst. Growth, 49, 132(1980), also
J. H. Neave, B. A. Joyce, P. L. Dobson, and N. Norton, Appl. Phys. , A31, 1(1983),
and references therein.
[2] Y. G. Chai and R. Chow, Appl. Phys. Lett. , 38, 796(1981).
[3] E. A. Wood, J. Appl. Phys. , 35, 1306(1963).
[4] A. Y. Cho, J. Appl. Phys. , 47, 2841(1976).
[5] H. Z. Chen, A. Ghaffari, H. Marko~, and A. Yariv, Appl. Phys. Lett., 51, 1094(1987).
[6] G. W. Wicks, W. I. Wang, C. E. C. Wood, L. F. Eastman, and L. Rathbun,
J. Appl. Phys. , 52, 5792(1981).
[7] W. T. Tsang, Appl. Phys. Lett. , 39, 134(1981).
[8] H. Marko~, T. J. Drummond, and R. Fisher, J. Appl. Phys., 52, 1030(1982).
[9] D. Z. Garbusov, I. N. Arsentyev, A. V. Ovchinnikov, and I. S. Tarasov, CLEO
procedings, THU4, 1988.
[10] D. C. Radulescu, G. W. Wicks, W. J. Schsff, A. R. Calawa, and L. F. Eastman,
J. Appl. Phys. , 62, 854(1987).
[11] P. N. Uppal, J. S. Ahearn, and J. W. Little, J. Vac. Sci. Technol. , B6, 597(1987).
[12] M. Tsuchiya, J.M. Gaines, R.H. Yan, R. J. Simes, P. 0. Holtz, L.A. Coldren and
P. M. Petroff, unpublished.
[13] T. Fujii, S. Yamakoshi, K. Nanbu, 0. Wada, and S. Hiyamizu, J. Vac. Sci. Technol.,
B2, 259(1984).

[14] A. Yariv, Quantum Electronics, 3rd Ed, John Wiley & Sons, New York, 1989.
[15] Loudon, The Quanfom Theory of Light, Oxford University Press, London, 1973.
[16] N. Ashcroft and N. Mermin, Solid State Physics, Holt-Saunders, Philadelphia,
1976.
[17] P. T. Landsberg, M. S. Abrahams, and M. Osinski, IEEE J. Quantum Elec-

-73-

tron. QE - 21, 24 (1985).
[18] B. Saint-Cricq, F. Lazes-Dupuy, and G. Vassilieff, IEEE J. Quantum Electron. QE22, 625 (1986).
[19] G. Lasher and F. Stern, Phys. Rev.

133, A553 (1964), and W. P. Dumke,

Phy. Rev. 132, 1998 (1963).
[20] A. Sugimura,IEEE J. Quantum Electron. QE- 20, 336 (1984).
[21] R. J. Nelson and R. G. Sobers, Appl. Phys. Lett. 49, 6103 (1978).
[22] J. Christen and D. Birnberg, Surface Science, 174,261 (1986).
[23] J.E. Fouquet and R. D. Burnham, IEEE J. Quantum Electron.

QE - 22, 1799

(1986).
[24] T. Matsusue and H. Sakaki, Appl. Phys. Lett. , 50, 1429(1987).
[25] M. G. H. Thomphoson, Physics of Semiconductor Laser Devices, John Wiley &
Sons, New York, 1980.
[26] Y. Arakawa and A. Yariv, J. Quantum Electron. QE - 22, 1887 (1986).
[27] H. Z. Chen,J. Paslaski,A. Yariv, H. Morkoc;, and G. Ji, unpublished.
[28] M. Asada, Y. Miyamoto, and Y. Suematsu, J. Quantum Electron. QE - 22, 1915
(1986).
[29] M. Asada, A. Kameyama, and Y. Suematsu, J. Quantum Electron. QE- 20, 745
(1984 ).
[30] E. Zielinski, H. Schweizer, S. Hausser, R. Stuber, M. H. Pilkuhn, and G. Weimann,IEEE
J. Quantum Electron. QE - 23, 969 (1987).
[31] S. P. Cheng, F. Brillouet, and P. Carree, J. Quantum Electron.

QE - 24, 2433

(1988).
[32] S. R. Chinn, P. S. Zory, and A. R. Reisinger, IEEE J. Quantum Electron. QE-24,
2191 (1988).
[33] S. Colak, R. Eppenga, and M. F. H. Schuurmans, IEEE J. Quantum Electron. QE-

-7423, 960 (1987).

[34] J. N. Schulman and Y-C. Chang, Phy. Rev. B31, 2056 (1985).
[35] D. Nino, M.A. Gell, and M. Jaros, J. Phys. C. 1986.
[36] G. Bastard, in Molecular Beam Epitaxy and Heterostructures, L. L. Chang and
K. Ploog, Eds. , Dordrecht, Boston, 1985.
[37] R. Eppenga, M. F. H. Schuurmans, and S. Colak, Proc. Int. Conf. Phy. Semi. ,
Stockholm, Sweden, 453 (1986).
[38] T. Hayakawa, T. Suyama, K. Takahashi, M. Kondo, S. Yamamoto, and T. Hijikata,
Appl. Phys. Lett. , 52, 339(1987).
[39] M. Hoenk, H. Z. Chen, K. Vahala, and A. Yariv, very recent results show strong luminescence from AlGaAs layer grown by MBE on etched (111) surface, Appl. Phys. Lett.,
54, 1347(1989).

[40] Casey and M. Panish, Heterostructure Lasers, Academic Press, New York, 1978.
[41] E. P. O'Reilly, K. C. Heasman, A. R. Adams and G. P. Witchlow, Superlattices
and Microstructures, 3, 99(1986).

-75-

Chapter 4
Potentialities and Lhnitations of GaAs-on-Si Technology

§ 4.1 An introduction

The amount of research on GaAs growth on Si by MEE and MOCVD has soared
world wide from near zero in 1984 to now include most major III-V research groups [l].
The potential applications have been so attractive that "everyone is trying it," so to
speak. In the beginning, most of the effort is directed at using GaAs for MESFET's and
HBT's in electronic circuitry [2]. Recently, the success of room temperature continuous
wave operation of GaAs quantum well lasers on Si has opened way to a entirely new
class of optoelectronic devices [3]. high-speed modulation of GaAs-on-Si lasers [3] and
high-speed GaAs-on-Si p-i-n photodiodes have also been reported for the first time [4].
In this chapter, some general questions concerning GaAs-on-Si research are answered. The advantages and limitations are discussed.

§ 4.2 GaAs versus Si

The large share of enthusiasm for GaAs and other III-V semiconductors is based
on their high electron mobilities and optoelectronic properties. III-V materials have
electron mobilities from 2 to 20 times greater than that of Si. More importantly, Si is

-76-

not a direct bandgap material suitable for optoelectronics.
In electronics, the search for physical limits to miniaturization reveals the reason
why GaAs and other III-V compounds have not replaced Si in integrated circuit applications: the limits have little to do with mobility; they are basically determined by (i)
avalanche breakdown fields, and (ii) thermal problems. It is necessary to have a large
bandgap to prevent intrinsic thermal excitation of carriers giving rise to a large intrinsic
conductivity. It is also favorable to have a large bandgap which allows the temperature
of a device to rise by a certain amount. On the other hand, high breakdown fields are
also associated with large bandgaps.
But the most important reason for the dominance of Si is the processing technology
developed around Si0 2, which can be formed on Si with remarkable ease and stability.
No III-V compound has a surface oxide layer that compares to Si02 as an insulator, as
a diffusion mask, and as a neutralizer of surface effects.
GaAs and III-V compounds therefore, can only have an impact in areas where Si
can not compete: their larger bandgaps, higher breakdown fields, and higher mobilities
make them attractive candidates for high performance microwave FETs, HBTs, lasers,
detectors, and novel heterojunction devices. In the lucrative digital electronics and
VLSI fields, however, GaAs is very unlikely to replace Si, since one or a few fast devices
can not improve system performance. The speed of a digital system is determined by
such things as delays in packaging (see Chapter 1). Low cost and high reproducibility
are essential to a large system, and for that reason Si will not be replaced easily.

§ 4.3 Advantages of GaAs-on-Si

Having been called a "system designer's dream" and a "material scientist's nightmare," GaAs-on-Si is both bitter and sweet: its promises are so attractive, yet its
problems so enormous. But first, we take a look at the positive side of it and list in

-77-

the following several obvious advantages of GaAs-on-Si technology based on today's
technology and our current understanding. There are certainly many more that one
can include.

Wafer size
GaAs wafers are currently available with diameters up to 3 inches.

For many

applications related to discrete devices, 3-inch wafers are large enough; however, LSI
logic and memory applications require larger chips. Monolithic microwave integrated
circuits (MMICs) are inherently large-area chips which should be processed on the
largest possible wafers. The task of developing large size GaAs wafers is under way;
however, the low thermal conductivity of GaAs imposes fundamental limitations on the
wafer parameter without sacrificing crystal perfection. Si on the other hand is available
in diameters up to 8 inches in high uniformity as well as purity.

Wafer cost
At today's price, a 2-inch GaAs wafer can cost about $250, while a Si wafer of
same size costs about $2, a difference of more than 100 times. For an epitaxially grown
GaAs wafer, however, the cost of growth is much higher than that of the substrate.
As a result, the use of GaAs-on-Si saves only a small fraction of the total wafer cost
after epitaxy. For some applications such as MESFETs, the devices can be fabricated
directly on a GaAs substrate by ion implantation without any epitaxial growth. In
these cases, GaAs-on-Si is much more expensive because of the high cost of epilayer
growth. As is often the case, significant savings of cost by using Si substrates can only
be realized when the GaAs-on-Si technology can benefit from the use of larger Si wafer
sizes unavailable in GaAs.

Wafer strength
GaAs is much more fragile compared to Si. A standard comparison of semiconductor hardness shows that Si is about 50% stronger than GaAs. Hardness is a rough

-78-

indication of the tendency of a wafer to break. A better measure is fracture strength.
An analysis of wafer fracture strength shows that Si is 2.5 times more resistant to
fracture. The handling losses during processing should be significantly reduced using
GaAs-on-Si material.

Wafer thermal conductivity
The thermal conductivity of Si is about 3 to 4 times better than that of GaAs. This
is an advantage for heat dissipation in the operations of high-power FETs and lasers.

§ 4.4 Limitations of GaAs-on-Si

Current limitations of GaAs-on-Si are mainly due to the poor quality of the crystal growth. It is understandable that the crystal quality of GaAs deposited on Si will
never be as good as single crystal GaAs.

Experience accumulated over the past 4

years, however, shows that crystal perfection has greatly exceeded previous expectations considering the large mismatch in both lattice constants and thermal expansion
coefficients. The quality of GaAs-011-Si has been steadily improved by new techniques
and even better material is expected in the future.

Thermal expansion mismatch
A serious problem for device application is wafer-bowing that results from different thermoelastic properties of GaAs and Si. On cooling from the epitaxial growth
temperature of 720°C, the free contraction of GaAs is 2.6 times greater than that of
Si. The postgrowth wafer is bowed with usually a concave GaAs surface and very high
tensile stresses. Depending on the growth technique, the wafer bow may range from an
acceptable 5 µm to greater than 50µm over a 2-inch wafer. In some cases as we have
observed, the tensile stresses exceed the elastic limits of GaAs and resulted in surface
cracks.
The extent of wafer bowing depends on the growth temperature, which should be

-79-

minimized whenever possible. For a typical laser growth, the substrate temperature is
as high as 720° C for Al GaAs, so alternative means have to be found to alleviate this
problem. Recently, according to some unpublished early reports, researchers at NTT in
Japan have used "modulated beam epitaxy" that deposits Ga and As layers alternately,
to grow bowing-free GaAs-on-Si. Additional techniques such as growth on patterned
substrate may be used to reduce bowing in future.

Process incompatibilities
Difficulties are encountered m the processing of GaAs-on-Si devices when wafer
cutting is required.

GaAs cleaves preferentially along (110) planes while Si cleaves

along (111) planes. A GaAs-on-Si laser, therefore will have additional problems due to
the cleaving of two facets. If the two facets are not perfectly parallel, photon loss due to
scatterings will be high. This problem, as will be shown later, can be minimized if the Si
substrate is lapped down to a very small thickness. Techniques called "microcleaving"
for on-wafer cleaving of GaAs lasers can be used to avoid the difficulty of cleaving Si
along (110) planes.
§ 4.5 Current status of discrete devices and integrated circuits

Over the past few years most kinds of discrete microwave devices have been demonstrated in GaAs-on-Si. The more successful ones are the devices whose operations involve majority carriers. For example, GaAs field effect transistors (FETs) on Si. FE Ts
are majority carrier devices and are not sensitive to defects in the crystal. This is why
FETs were among the first devices fabricated in GaAs-on-Si with performances comparable to similar devices on GaAs substrates. They have the advantage of reduced wafer
breakage due to the more robust Si wafer, reduction in substrate cost, and increased
wafer size and yield. The best GaAs FET on Si has a 360mS/mm transconductance
for a 0.25µm gate length, 55 GHz cutoff frequency, and a noise figure of 2.8dB at 18
GHz [5]. This is comparable to the best data on GaAs substrates.

-80-

Other such devices like modulation-doped field effect transistors (MODFETs) also
known as HEMTs have also been reported with transconductances of 170ms/mm at
room temperature and 275ms/mm at 77K for a lµm gate length, and a gain cutoff
frequencies as high as 23 GHz [6].
Heterojunction bipolar transistors (HBTs) are minority earner devices, they are
very sensitive to material defects. The first reported GaAs bipolar transistor on Si
showed a current gain of 10, and current densities up to 105 kA/ cm 2 , and very good
performances at microwave frequencies as high as 40 GHz have been reported [7]. Light
emitting diodes (LEDs) are another minority carrier device that have been reported

[8].
Solar cells [9] and focal plane arrays [10], as well as ring oscillators [11] have attracted much attention and have been reported.

Lasers and optoelectronic integrated circuit applications
They are the main topics of the next chapter. Major breakthroughs and advances
have been made in the past two years; many happened in this laboratory.

§ 4.6 Conclusions

In this chapter, we have presented both advantages and limitations of GaAs-onSi technology based the limited experience, and summarized the short history this
young research field has had. The GaAs-on-Si technology has already been used to
fabricate most common microwave and optoelectronic devices with rather surprising
successes. Future research in this field will overcome current technological barriers and
push this promising technology to a state-of-the-art production stage for many exciting
applications.

-81§ 4.7 References

[1] R. Houdre and H. Morko[2] T. Nonaka, M. Akiyama, Y. Kawarada, and K. Kaminishi, Japn. J. Appl. Phys. ,
23, 1919(1984 ).

[3] H. Z. Chen, J. Paslaski, A. Yariv, and H. MorkoJan. , 61(1988).
[4] J. Paslaski, IL Z. Chen, H. Marko<;, and A. Yariv, Appl. Phys. Lett., 52, 1410(1988).
[5] R. J. Fisher, N. Chand, W. F. Kopp, C. K. Peng, H. Marko<;, K. R. Gleason, and
D. Scheitlin, IEEE Trans. Electron Devices, ED - 33, 206(1986).
[6] R. J. Fisher, T. Henderson, J. klem, W. T. Masselink, W. F. Kopp, H. Morlrnand
C. W. Litton, Electron. Lett. , 20, 945(1984).
[7] R. J. Fisher, N. Chand, W. F. Kopp, H. MorkoAppl. Phys. Lett. , 49, 397(1985).
[8] R. M. Fletcher, D. K. Wagner, and J.M. Ballantyne, Mater. Res. Soc. Symp. Proc.,
25, 417(1984).

[9] B-Y. Tsaur, J. C. C. Fan, G. W. Turner, F. M. Davis, and R. P. Gale,
Proc. IEEE. Photovolt. Spec. Conf. , 1143(1982); J. C. C. Fan, B-Y. Tsaur, and
B. J. Palm,
Proc. 16th IEEE Photovolt. Spec. Conf. , 692(1982).
[10] R. C. Bean, K. R. Zanio, K. A. Hay, J. M. Wright, E. J. SAller, R. Fisher, and

H. Morko[11] H. Shichijo, J. W. Lee, W. V. McLevige, and A.H. Taddiken, Proceed. Int. Symp. GaAs
and Related Compounds, Inst. Phys. Conf. Ser. , 83, 489(1987).

-82-

Chapter 5
MBE Growth of GaAs-on-Si Quantu1n Well Lasers

§ 5.1 An introduction

The most attractive application of GaAs-on-Si material is the integration of the
high-speed and optoelectronic properties of GaAs and the state-of-the-art processing
technology of Si VLSI on a single chip. The high-speed on-chip GaAs lasers and detectors can increase the signal fan-in and fan-out ( see Chapter 1) and interconnect a
complex multi-chip supercomputer system.
The key to GaAs-on-Si optoelectronics is the quality of GaAs grown on Si. Currently, the quality of GaAs-on-Si is still low. A typical GaAs-on-Si crystal has a defect
density of N

:::::J

10 6 cm- 2 and surface tension of (j

:::::J

108 dyn cm- 2 , mainly due to the

lattice constant mismatch between GaAs and Si. The high density of defects does not
affect the performance of majority carrier devices such as FETs; however, it drastically
degrades the performance of minority carrier devices such as lasers. For nearly 4 years,
researchers worldwide have been working intensely to improve the electronic and optical quality of GaAs-on-Si, with the eventual goal of achieving the room temperature
continuous wave (CW) operation of a GaAs-on-Si laser. The technological barriers have
been the high density of defects in GaAs grown on Si. In this chapter, we will present

-83-

the MBE growth of a single quantum well GRINSCH GaAs/ AlGaAs laser on Si that
achieved the world's first room temperature CW operation. Difficulties of GaAs-on-Si
growth are discussed. Methods employed to overcome them are described in detail.
After the CW operation, stripe geometry GaAs-on-Si lasers were fabricated and modulated by high frequency microwave signals. High-speed GaAs-on-Si p-i-n detectors were
also grown with a similar procedure.

§ 5.2 Special problems associated with GaAs-on-Si growth

When a polar semiconductor (GaAs) is grown on a nonpolar semiconductor (Si),
special problems arise which are not present in the conventional polar-on-polar heteroepitaxy growth.

They are: (i) the problem of antiphase disorder on the polar

(GaAs) side of the interface, (ii) the lack of electrical neutrality at the interface, and
(iii) the problem of cross-doping. In addition, the large difference in thermal expansion
coefficient and lattice constant mismatch which are not related to polar-on-nonpolar
semiconductor growth can cause other severe problems.

Antiphase disorder
Si has a diamond structure which consists of two interpenetrating face-centered
cubic Bravais lattices. GaAs has a similar crystal structure, except the two face-centered
cubic lattices are not the same; the first one is occupied by Ga atoms, and second one
by As atoms. Because the Si lattice is occupied by only one type of atom, the crystal is
invariant to a 1r/2 rotation along (100) and the (011) and (Oll) directions are equivalent.
However, this is not so for GaAs. Therefore, when the growth of GaAs-on-Si is started,
the orientation of GaAs ( there are two orientations possible) is not completely defined
for the entire wafer but may be defined locally in some small domains which are known
as antiphase domains (APDs). Antiphase boundaries (APBs) occur when these small
regions with defined orientations 1r/2 different, are joined (Figure 5.1).

-84-

(a)

APO

(b)

Figure 5.1 Schematic drawing the antiphase domain boundary formed by two regions with (a)
an As prelayer and (b) a Ga prelayer.

Microscopically, on an exact (100) Si substrate, there is no distinction between Ga
and As sites and the first monolayer may be randomly occupied by Ga and As atoms. A
(100) GaAs is a crystal with alternating Ga and As planes. Therefore, the growth may
begin in some regions of the interface with a Ga plane, and in others with an As plane.
When this happens randomly, we get high defects of anti phase boundary disorder.

Electrical neutrality
\Vhen both types of domains meet, massive Ga-Ga ( or As-As) bonds are created,
which are electrically charged and can act as acceptors ( or donors).

-85-

Step-doubling and substrate misorientation
Moreover, a real (100) Si surface will always contain atomic steps, and the steps
with an odd number of atomic layers will confuse the distinction between Ga and As
sublattices and then induce APD even if the first atomic plane is occupied by only one
type of atom. Fortunately, most steps on (100) Si surface appear to be double steps [4]
after the substrate is heated to 1100°C before the growth is started.

APD free growth of GaAs-on-Si
Numerous studies [3,5,6] indicate that APD is the result of (i) a nonuniform coverage of the first monolayer, and (ii) single ( odd number) steps at the Si surface.
Therefore, it has been proposed [5] that (i) a uniform coverage with an As ( or Ga)
prelayer be used, and (ii) tilted substrates which are known to have double steps be
used. Using these techniques, APD free (100) GaAs has been grown [6]. All of our
laser structures have been grown this way. Qualitatively, an arsenic prelayer helps the
crystal define its orientation while double steps eliminate possible confusion resulting
from odd numbers of steps. But even if the above conditions cannot be satisfied everywhere on (100) Si surface at the beginning of growth, a subsequent slow growth of
a transition layer will allow different islands to form, which will intercross and finally
grow into one bulk material. There is still, however, no agreement among experts on
a quantitative microscopic theory that can explain the strange behavior of transition
layer growth which strongly depends on several parameters [7].

Thermal expansion mismatch
The problem with the difference m thermal expans10n coefficients is of intrinsic
nature and the only thing we can do is to reduce its effects. One obvious scheme is
called "selective area growth," where GaAs is grown on certain isolated areas so that
thermal mismatch can be confined to small isolated regions instead of entire wafer.
Gradual cooling after the growth is finished is also important. Samples suddenly cooled

-86-

from 720° C to room temperature often have surface cracks visible under an optical
microscope. This problem is potentially serious for long lifetime optoelectronic devices
such as lasers and has not been studied carefully.

Lattice constant mismatch
The 4.1% lattice mismatch requires one type I dislocation for every 25 atomic
planes and one type II dislocation for every 18 atomic planes ( type I,II dislocations
are discussed later). This results in a dislocation density of 10 12 cm- 2 , which is much
higher than the practical limit of 104 cm- 2 which devices can tolerate. To find effective
techniques for dislocation density reduction, we must have a good understanding of the
nature of these defects. Generally, it is known that there are two ways an epitaxial layer
can accommodate a lattice mismatch with the substrate: by introducing strains, and
by introducing dislocations. In the first case, the lattice mismatch is accommodated
by an elastic deformation of the lattice; however, for GaAs-on-Si system, the critical
thickness of GaAs layer is 50 A [2], and beyond which, the strain energy which is
proportional to the thickness of the layer becomes larger than the minimum energy to
generate dislocations. In practice, we are always in the second case: dislocations have
to be generated to account for the large lattice mismatch. Therefore, the best we can

do is to confine these dislocations to the GaAs/Si interface so that they will not affect
the device operations several microns away from the interface.
Several popular methods have been used to reduce the effect of dislocations. Their
main goal is to confine mismatch dislocations to the interface and prevent them from
transforming to threading dislocations that will travel through the later growth.

Substrate misorientation
The two types of misfit dislocations type I and type II that we mentioned earlier have been studied experimentally [8]. The main conclusions are that the type I
dislocations are inactive sources for generation of threading dislocations because their

-87-

Burgers vectors lie in the (100) plane and are parallel to (011) and (011) directions,
and the type II dislocations on the other hand, can move through the crystal by gliding
along the ( 110) planes since their Burgers vectors are inclined from the interface by
45° (Figure 5.2 ). The type II dislocations are therefore highly active sources for the
generation of threading dislocations which are harmful to device operations. Therefore
the goal is to increase the fraction of type I dislocations in the misfit dislocation network and effectively reduce the fraction of type II active sources for the generation of
threading dislocations.
In practice, this is accomplished by using a substrate tilt. It has been known [8]
that steps at Si surface preferentially induce type I dislocations.

Thus it has been

suggested [5,9] that a tilted substrate with one step every 25 atomic planes be used.
The angle of such tilt is 1.6° for (100) tilted toward (011), and (011) being the direction
that the steps run along. If the substrate is tilted toward (001) then the angle would be
2.3°. Figure 5.3 shows a staircased Si surface with atomic steps. It has been proposed
and demonstrated [5,9] that a slightly larger substrate tilt, 4° from (100) toward (011),
is more effective. This is because on a real Si surface, the steps do not occur at regular
intervals. The slightly larger tilt can make sure that there are very few intervals between
two steps larger than 25 atomic planes. Using this technique, dislocation density has
been greatly reduced to as low as 10 5 cm- 2 to the mid-10 4 cm- 2 range. This technique
has been used throughout our research on GaAs-on-Si devices.

Strained layer superlattice
GaAs/InGaAs strain layer superlattice (SLS) structures have been successfully used
to bend the dislocations [5,9]. The direction of the bending depends on the sign of strain.
The use of a material with a larger lattice constant will induce a compressive strain in
this layer, which will tend to repulse the dislocations from the strain superlattice.
We have experimented with GaAs/InGaAs strain layer superlattice in the buffer

-88-

(a)

oGa

•As
(b)

Figure 5.2 Schematic drawing of ( a) the type I misfit dislocation with the Burgers vector parallel
to (100) and (b) the type II misfit dislocation with the Burgers vector inclined from (100) by an
angle of 45°.

layer. Thus far, the results are inconclusive.

Rapid thermal annealing
Once the laser structure is grown, ex-situ annealing can be performed. If done
properly, it can reduce the density of dislocations dramatically. Annealing experiments
have been done by several groups and they all report an improvement in the quality
of GaAs epitaxial layer [10]. The ex-situ thermal annealing performed on our samples
are done at 850° C under an inert gas. The annealed lasers showed no delay in the light

-89-

(100)

Tilt angle with
respect to the
(100) plane

Figure 5.3 Schematic drawing of a (100) substrate tilted toward (011).

response when pumped with electrical pulses ( usually a turn-on delay typically of 10-50
ns is observed indicating the presence of defects which must be saturated first by the
initial pump pulse). This is a noticeable improvement over the unannealed samples for
which a delay of light response is common.

§ 5.3 Substrate preparation

Before GaAs can be grown on Si, a clean Si surface in a ultra-high vacuum MBE
environment is required. Classical chemical cleaning techniques of Si often produce
a surface with carbon (C) and oxygen (0) contaminants. The oxygen on the surface
can be thermally removed by heating the Si substrate to 900 - 1000°C, but the car-

-90-

bon can only be removed partially at 1200°C, which is too high for a standard MBE
system where the maximum substrate temperature is limited to 900°C. Moreover, the
carbon may react with the Si substrate at 800°C forming SiC. The presence of SiC on
the surface is known to have a catastrophic effect on epitaxial growth on Si. These
considerations require a surface preparation procedure that can remove the carbon on
the surface, and can produce a thin protective layer that can easily be removed inside
MBE. This protective layer can also prevent carbon contamination by air during the
loading process.
The cleaning procedure we have been using was invented by R.C. Henderson [1]
and modified by H. Morkot; [2]: (i) degreasing and removal of a native oxide layer on
the surface, (ii) several iterations of growing and stripping of a chemical oxide layer.
During this step, any contamination (including C) at the surface will be buried in the
oxide layer which is then removed by a subsequent stripping, and (iii) the growth of
a thin and volatile oxide layer that will be easily desor bed by heating the substrate
inside the MBE's growth chamber. The details of this cleaning procedure are listed in
Appendix V.

After the cleaning, the substrate is mounted onto a homemade Mo block for direct
radiation heating. The mounted substrate is then loaded into the loading chamber of
the MBE system where the substrate is gradually heated to 900-1000°C for 2 minutes for
oxide desorption. Next, the substrate is transferred to the growth chamber of the MBE
system, and the substrate is heated up to about 900°C again, to anneal the surface
damage and blow off any impurity condensation on the surface during the transfer.
Finally, the substrate is cooled down to an appropriate initial growth temperature to
be discussed later. After this oxide desorption and thermal treatment, a clear (2x 1)
RHEED pattern can be observed, which indicates a clean Si surface with a standard
reconstruction.

-91-

§ 5.4 Transition layer growth

First, to prevent the antiphase domains when the growth is attempted on a (100)
Si surface, it is started with an As preexposure. This is done at the beginning of the
growth by opening the As shutter for a sufficiently long time until the Si surface is
completely covered by a few monolayers of As atoms. Should a Ga prelayer be used?
No. This is because an As prelayer is energetically favored. If a Ga prelayer is put on
Si, then As atoms next deposited on top of Ga would always get under the Ga layer
and make bonds to Si.
After the preexposure is done, the substrate is exposed to both Ga and As fluxes
and the normal growth is started. The As pressure for preexposure as well as for growth
should be PAs 4

= l x 10- 7 torr using a conventional cell, or PAs = l x 10-s torr using

a cracker cell. At this pressure, a 5 second preexposure of As is enough. It has been
discovered experimentally that As 2 has a higher sticking coefficient to Si surface, and
it can tolerate a higher starting substrate temperature. According to our extensive
experience, to obtain a As prelayer, a temperature lower than 200°C is required for an
As 4 source and a temperature as high as 400°C can be used for an As 2 cracker cell.
The growth rate of GaAs in the very beginning is kept lower than 0.lµm/h since
the newly grown crystal needs some time to find the best arrangement, or to minimize
the free energy, so to speak. A very important trick is that whenever the GaAs growth
becomes bad on the RHEED screen, Ga cell temperature should be lowered to let the
growth recover. This has been used often in the early stage of a GaAs-on-Si growth.
During this transition layer growth, the RHEED pattern will change from very
spotty to streaky. A spotty pattern ( typically seen in small angle X-ray scatterings of
powdered 3D specimen) indicates a three-dimensional type of growth, and a streaky
RHEED pattern (which represents a 2D surface reconstruction, see Chapter 4) is associated with a standard two-dimensional GaAs surface reconstruction. The substrate

-92-

temperature during this growth is gradually raised to a standard growth temperature of
580°C which normally gives a lµm/h GaAs growth rate. To obtain good surface morphology, the time of the transition layer growth can be as long as two hours although
this period should be as short as possible, since low temperature growth of GaAs is
known to have poor optical and especially poor electrical quality. There is a tradeoff
between the arsenic sticking coefficient and the quality of the transition layer: a low
starting temperature is required for a good arsenic sticking on Si substrate, and a high
growth temperature is needed for high crystal quality. To accomplish both, the growth
should be initiated at a low substrate temperature at about 200°C, and immediately
after the first 1000 A of growth, the substrate temperature is raised to 580°C. The cell
temperatures of Ga and Si are also raised, gradually following the increase of substrate
temperature. When the substrate reaches 580°C, the Ga cell temperature is set to what
should give a lµm/h growth rate.
The thickness of the transition layer grown at a slow rate and at a low substrate
temperature is usually between 250-2500 A. This is a layer of amorphous material which
can be annealed to become crystalline (hopefully) at 580° C. Because this transition layer
is of low quality, the defect density is usually high. As a result, the doping level in this
layer can be effectively reduced as we have experienced. What should be an n+ GaAs
doping level has produced rather low doping and becomes n-. Furthermore, a large
voltage drop across this layer sometimes as large as 10 V, compared to an expected 1.5
V has been observed in current-voltage (I-V) measurement. Use of a "cracker cell" can
help solve this problem since the starting growth temperature is above 400°C and can
quickly be raised to 580°C, thus making the transition layer very thin and the voltage
drop very small. Also, the quality of GaAs grown at Ts between 400 - 580°C is much
better than material grown between 200 - 580° C.
At this point, there seem to be three key elements to a successful transition layer

-93-

growth: (i) a sufficiently long As exposure before Ga shutter is opened, (ii) gradual
rise in Ga and dopant cell temperature, and (iii) bringing the substrate temperature to
600° C as soon as possible.

§ 5.5 Room temperature CW operation of GaAs-on-Si lasers

It has been long established that room temperature continuous wave (CW) oper-

ation of a GaAs-on-Si current injection laser is a very important step toward eventual
realization of commercial OEICs of GaAs-on-Si. For several years, researchers worldwide have been actively seeking ways to obtain at first low-threshold pulsed operations
of GaAs-on-Si lasers, and eventually CW operations. What is described in the following
is an account of the most important research contribution of this thesis project.

First step: low-threshold current density pulsed operations
By mid-1987, the lowest threshold current density obtained in a GaAs-on-Si laser
under pulsed operation was 3500A/cm 2 [11]. This was mainly due to the poor quality of
epitaxial growth of GaAs-on-Si. It was obvious that a significant reduction was needed
to achieve room temperature continuous wave operations.
Before the work of GaAs-on-Si was started, the growth conditions of GaAs-onGaAs lasers had been optimized by us to such an extent that we could successfully
grow GaAs-on-GaAs GRINSCH lasers with threshold current densities as low as 140
A/cm 2 regularly.
The work of GaAs-on-Si, however, was more complicated. First, we had to modify
the Mo blocks used for indium mounting for direct heating of Si substrates, because
indium-free mounting allows a quick flash of Si substrate at 900° C for oxide desorption,
which can not be done with indium mounting (indium evaporates at 900°C and the
substrate will separate from the mounting block and drop). This was done by opening
a large hole in the middle of a conventional Mo block, of roughly the size of a 2 inch

-94-

Si wafer

Removable
Pin

♦ heating+

Figure 5.4 The indium-free mounting used in the GaAs-on-Si growth.

Si wafer. Three pins made of tantalum wires were used to clip down the wafer to the
now O-ring-shaped Mo block (Figure 5.4), with one of them removable after each use.
Inside the MBE's growth chamber, the heating filaments on the sample manipulator can radiate directly toward a Si substrate. This heating method has the distinct
advantage over the conventional In mounting heating method where heat is transferred
by liquid In on contact, and as a result, the substrate temperature can be changed
rapidly.

It was pointed out m Chapter 4 that the quality of AlGaAs growth depends on

-95-

the substrate temperature, preferably at 720°C. With direct heating, this can easily
be accomplished since it takes less than 10 seconds to raise the substrate temperature
from 600°C to 720°C, compared with In mounting which requires as long as 2 minutes
due to the large thermal capacity of the Mo block material plus an In layer between
the heating pads and the substrate.

This heating scheme is not only essential for

GaAs-on-Si growth, but also good for GaAs-on-GaAs growth.
The starting growth tern perature for the first few successful low-threshold lasers
was about 300-400°C, which was followed by a 5-second As preexposure. These lasers
were used to demonstrate low-threshold current density, and one of them was used to
perform CW operation at room temperature.
Later, we have found that with an As 4 source and a lower starting temperature
(200°C), an O-ring on the edge of wafer (Figure 5.5) can be observed. This is because
the direct heating allows the middle portion of the wafer facing the heating pads to
cool down quickly to 200°C, while the area on the outside not facing the heating pads
cools down slowly with the Mo block. Since As atoms only stick to the cooled area ( at
200°C), an O-ring is formed which is the boundary between the As-prelayered region
in the middle and the ( Ga,As )-mixture prelayered region on the outside. The GaAs
inside O-ring is also APD free and outside O-ring is APD prone. With an As 2 source,
however, such an O-ring is not observed since As 2 can stick to a surface at a higher
temperature. All of the successful GaAs-on-Si wafers grown had the above feature.
Furthermore, the lowest threshold current density reported prior to our work had
been 3500 A/cm2, and our lasers achieved a threshold current density as low as 600
A/cm 2 [12] almost 6 times lower. Figure 5.6 shows the light vs. current relationship
of lasers made using two cavity lengths. Curve A represents a (120µm x 520µm) laser
with Ith

= 0.6A and curve B represents a (120µm x 1210µm) laser with Ith = 0.87A. A

threshold current density of Jf/, ~ 600A/cm 2 was obtained, which indicated for the first

-96-

As does not stick because
this region cools off slowly
(surface temperature is
higher than the center).

. . . . . . .. . . . . . . . . . . ... ... .... ... . . . ..

APO t_ree GaAs
(100)

..............
. . .. . ... . .. . .. ..... . ..... .. ... .. .... . ... . ... ....... ... ... ........ ......... ... .. .... ....... ............
.. ...... ....... .. ........... ........... . .. ..... .... ..... . ... .. .. . . . ..

Figure 5.5 The effect of growth temperature on As sticking coefficient and surface morphology
shown on a finished growth. The center region is APD free and has good morphology but the
outside ring contains a high density of APD related defects.

time that high device-quality Ga.As can be grown on a Si substrate.
Until this time we did not seriously think that C\V operation was possible. There

-97-

80

70

Cl)

.........

60

E 50

...

(l)

40

0...

--a. 30
::,

::,

20
10

0.7

0.8

0.9

1.0

1 .1

1.2

1.3

1.4

1.5

Current (A)

Figure 5.6 The light versus current characteristics for two low-threshold GaAs-on-Si broad area
lasers. Both exhibited threshold current density of 600 A/cm 2 •

were, however, some other very difficult technical problems to overcome even after
achieving a low threshold current.

Second step: Room temperature continuous wave operation
Several very good GaAs-on-Si samples were grown by l'vfBE shortly afterwards.

-98-

Threshold current densities as low as 214 A/cm 2 were obtained from one of them.
Historically, the first GaAs-on-GaAs CW laser had a threshold current density of
l.6x 10 3 A/cm 2 [13) which is several times higher than our GaAs-on-Si result. A CW
operation of GaAs-on-Si laser thus appeared possible.
The first question is what kind oflaser device should be attempted in CW operation.
A laser with proton-bombardment-defined stripe was used in the first CW operation of
GaAs-on-GaAs lasers. Such a device is easy to fabricate and was chosen for our first
attempt. Unfortunately, we were unable to achieve any CW operation. The failure
was mainly due to the damage incurred during device processing, and the difficulty in
making good thermal contact to a heatsink. In addition, the stripe laser exhibited a
much higher threshold current density than that of the broad area devices, and as a
result, the heat dissipation per unit area is much higher. Therefore, we decided that
a broad area laser would be used to accomplish CW operation. The difficulty with a
broad area laser, is that we have to make uniform thermal contact over the entire area.
Having decided the type of laser device to be used, we focused our attention on
other technical problems.

First, unless we could solve the problem of lapping and

cleaving we would not obtain low-threshold lasers for CW operations. The problem
with lapping occurs when a wafer is being removed from a lapping block. The removal
involves heating up the lapping block to melt the wax used to attach the wafer. This will
cause a large wafer-bow due to the large difference in thermal expansion coefficients. In
addition, the difference in thickness between GaAs epilayer ( a few µm) and Si substrate
( about 100 µm thick) causes bowing of the entire wafer visible by the naked eye. When
this happens, the two cleaved facets are not parallel and the photon loss in the cavity is
very high. In the worst case, the wafer is bowed to a degree that further processing is
impossible. The solution was to soak the entire lapping block with the wafer in acetone
solution, so that the wafer could come off the block by itself with the disassociation of

-99-

wax. Next, we were faced with the problem of cleaving, which is caused by the fact
that GaAs cleaves along (110) and Si cleaves along (111 ). To obtain a good alignment
of two cleaved facets, it is important to lap the wafer as thin as possible. The cleaving
should be done in such a way that the knife-edge used in cleaving is as close to the
center of the piece of wafer as possible for a symmetric breakage.
After the fabrication, we were faced with the very difficult problem of mounting
the laser with the substrate side up on a heatsink for heat dissipation through thermal
contact (the distance between the p-n junction and the heat sink is only 2.5 µm compared to 100 µm in the substrate down mounting). At first, lapped copper blocks were
used, but this did not work because the copper blocks used were not smooth enough
and copper was not sufficiently good a heat conductor. At one point, the laser did
operate CW, but the lifetime was not long enough for recording.
Finally heatsink squares made of industrial diamond were ordered whose thermal
conductivity is 3 times better than that of copper. The mounting scheme was the
following: (i) mount a piece of diamond on a copper block using an indium alloy that
has a high melting point around 400°C, and (ii) then mount the laser chip upside down
on the diamond with a low melting point indium alloy around 200° C. The amount of
indium on the diamond surface is important since excessive indium will flow and short
the p-n junction of the laser which is only 2.5 µm away from the mounting surface. Too
little indium will leave gaps in the indium underneath the laser where heat can not be
removed and can cause burning when CW operation is attempted.
The actual attempt to obtain CW operation was hampered by the unsuccessful
effort in mounting. This tedious task lasted about three weeks without success, until
one day when everything was just right and we finally achieved the first CW operation.
The first CW operation is shown in Figure 5.7, which was operated for 5 minutes.
The operation was stopped to preserve this first CW laser as a souvenir. A light-current

-100-

characteristic is shown in Figure 5.8. The laser that was operated CW has an surface
area of 120x 980 µm 2 and has a threshold current of 350 mA. It lased at a wavelength
of 8630 A. This was the first CW operation ever obtained, and it was reported at the
GaAs-on-Si Workshop in Marina del Rey on June, 18, 1987, in California [14]. Until this
time no current-injected room temperature CW operation had been reported, although
there has been report of optically pumped CW operation without a p-n junction.
Later, we improved and perfected the mounting method (Figure 5.9). Its details
are given in the following: first, mount the diamond on the copper block as described
before; then put a drop of rosin-based solder flux on the diamond, the size of the flux
being large enough to cover the entire diamond. Next, a very small indium ball (they
are commercially available from Indium Corporation of America) is dropped into the
flux and placed at the center of the diamond heatsink. A cleaved single laser can then
be dropped onto the flux carefully. It should stay on the center top of the flux. The
entire (laser chip + diamond heatsink + copper block) stack is then heated up slowly.
The flux will become more fluid and the laser chip will come down slowly as a result of
gravity. The laser makes contact to the indium ball sitting inside the flux, and when the
temperature is high enough, the indium ball melts into a small pool covering the entire
diamond surface. Down with the indium ball comes the laser chip onto the diamond,
which makes a good thermal contact to the diamond through a very thin layer of In.
No gap exists between the chip and diamond since there is no air in the flux drop.
The entire mounted block is then removed from heat. After it cools down, it is put in
acetone solution to remove the flux on the facets (mirrors).
After the success of CW operation, several other characteristics were measured to
obtain a complete picture of the GaAs-on-Si laser.

Polarization of laser emission
It has been speculated ( unpublished results among some researchers) that the po-

-101-

Figure 5. 7 A photograph of the first room temperature, current-injected, continuous wave (CW)
operation in a broad area GaAs-on-Si laser. The center frequency is 8630 A,and the horizontal
scale is 100 A. The quantum well width Lz is 125 A. It was operated CVV for 5 minutes and had
a threshold current of 350 mA for an area of 120 x 980µm 2 .

larization of light from a GaAs-on-Si laser consists of both transverse electric (TE) and
transverse magnetic (TM) field modes due to the strain induced by the residual lattice
and thermal mismatch. To investigate this, we have measured the polarization of light

-102-

120µ

~---~

"'"'
1w

1900µ

(.)

Lt
.......

Jth J::$ 263 A/cm 2 from peak

~ 100

Jth s:::s 322 A/cm 2 from average

'-"

J th ~ 214 A/ cm 2 from optical spectrum

a:

80

Cl.

40

~ 60
20
0.4 0.5

0.6 0.7

0.8 0.9 1.0

1.1

1.2

1.3

1.4

CURRENT (A)

Figure 5.8 The light versus current curve of a CTV broad area GaAs-on-Si laser that exhibited

the lowest threshold current density of 214 A/cm2 .

from two lasers with very large thermal and lattice mismatch that resulted in some
visible surface cracks. In one case, the cracks were running perpendicular to the laser
cavity, in the other case, the cracks were running parallel to the cavity. A polarizer was
inserted between the laser and a monochrometer which can pass either one of the two
polarizations at a time. The polarization resolved spectra for both lasers are shown in
Figure 5.10 and Figure 5.11. The measurement did not confirm the existence of any

TM mode.

-103-

Single piece of
laser with its
upside down

In alloy melts
at 400'C

F1ux will
evaporate
at20Cl'C

Gravity brings

down the laser
and heat melts
In while flux
is pushed out

In alloy melts
at 160'C

Diamond

450'C

25"C

2oo·c

Figure 5.9 A better mounting method of a laser onto a heatsink. This is used in industry presently.

This study, however, shed some light on the apparent effect of strain on mode
selection. In the laser with surface cracks perpendicular to the cavity, the spectral lines
are centered around 8700 A( 1.425 e V ), much like a typical GaAs-on-GaAs laser except
some very weak modes at higher energy around 1.48 eV. In the laser with surface cracks
running parallel to the cavity, the spectrum is much different and consists of peaks of
com parable strength centered around 1.425 e V and 1.48 e V. All of the above are TE
modes. This is consistent with the fact that the mirror reflectivity of TE polarized
light is always higher than that of TM polarized, because the boundary conditions
imply that the light polarized perpendicular to the plane of incidence (TE) always has
higher reflectivity. A TE mode therefore, suffers lower mirror losses. At this time, we
attribute the observed difference in the lasing spectrum to the effect of strain on the
gain coefficient [15].

-104-

1.38eV

1.55eV

550 +2, Ucracks

1.42eV

1•225mA

1.50eV

8000

9000(A)

Figure 5.10 The optical spectrum obtained for a GaAs-on-Si stripe laser with surface cracks
parallel to laser cavity.

Near and far field patterns
Near and far field patterns can provide valuable information of the spatial modes
in a laser. It is well known that filamentation can cause undesired instabilities in the
device performance. Figure 5.12 shows the near field of a typical ridge-waveguide laser,
which clearly demonstrates good current guiding in the lateral dimension and a single
lasing filament.
A plot of light vs. current also shows no sign of kinks typically caused by mode

-105-

1.38eV

1.55eV
1.42eV

550 +1,icracks
1.426eV

I=180mA

1.48eV

. I
8000

9000(A}

Figure 5.11 The optical spectrum obtained for a GaAs-on-Si stripe laser with surface cracks
perpendicular to laser cavity.

hoping. A very narrow and single lobed far field pattern (Figure 5.13) is also observed.
The beam angle (in the lateral dimension) of 4.8° compares very well with that of
typical GaAs-on-GaAs stripe lasers [15].

§ 5.6 Ridge-waveguide geometry stripe lasers

The response of a device is determined by the RC time constant in any high-speed
microwave modulation measurement. In the case of a laser, the capacitance of the p-n
junction can severely limit the speed of modulation. Stripe geometry lasers with small

-106-

NEAR FIELD PATTERN
I= 180mA

7.5µm

50µm

Figure 5.12 The near field pattern of a 10 µm GaAs-on-Si stripe laser which is biased at three
times above its threshold.

junction areas (capacitances) have been used for high-speed modulation experiments.
There are numerous structures of stripe geometry lasers published in literature,
which can be put in two groups: (i) gain guided, and (ii) index guided. The index
guided structures are usually fabricated by liquid phase epitaxy regrowth. The gain
guided structures are based on the confinement of carriers by finite diffusion length.
For example, if the stripe width is 1 µm, then the total current spread should be less
than 10 µm at the junction. This current spread can be further reduced if a ridge

-107-

FAR FIELD PATTERN
I= 180mA

32°

Figure 5.13 Tl1e lateral profile of the far field of a 10 µm GaAs-on-Si stripe laser which is biased
at three times above its threshold.

is etched. Gain guided structures with a etched ridge are call ridge-waveguide stripe
lasers. These lasers do not have threshold currents as low as the index guided ones, but
they are easy to fabricate and can be obtained in large quantities.
This procedure (Appendix VI) is the one we have been using in this thesis work
and it has the 10 steps.

§ 5.7 High-speed modulation of GaAs-on-Si stripe lasers
One of the most important goals in GaAs-on-Si research is to modulate GaAs lasers

-108-

on Si substrate using microwave signals that can be generated from a Si VLSI chip,
or from a GaAs chip on the same Si wafer. Such modulation takes advantage of the
optoelectronic capability of GaAs and the state-of-the-art of Si VLSI technology in
future supercomputer systems.
The stripe laser used in modulation experiments is shown in Figure 5.14, and the
experimental set-up is shown in Figure 5.15. The stripe laser is mounted on a copper
heatsink which is screwed onto a microwave package available from Ortel Corporation.
The DC current bias and the microwave signal are applied through a microwave biasTee. The light from the laser is collected and collimated by a microscope objective,
and then focused onto a high-speed photodetector (3dB bandwidth = 8GHz ). The
output of the photodetector is sent to a spectrum analyzer. The frequency response
measurement is shown in Figure 5.16. As can be seen, the 10µmx380µm stripe laser
exhibited a threshold current of 40 mA, a lasing wavelength of 8650 A, and a modulation
corner frequency of 2.5 GHz. The modulation was performed at frequencies as high as
4.5 GHz (15,16).
This result compares very favorably with those obtained from GaAs-on-GaAs lasers
of similar structures (they have a corner frequency of about 2 GHz [17]). Furthermore,
such a laser can be used in as chip-to-chip optical link together with a high-speed
GaAs-on-Si detector.

§ 5 .8 High-speed GaAs-on-Si p-i-n detectors

Si is not a good material for detecting GaAs laser emission at high-speed since Si has
a large absorption depth of nearly l0µm and thus a long sweep-out time proportional
to this length. GaAs on the other hand, has only a absorption depth of about lµm and
a carrier speed two times higher. As a result, GaAs is expected to out perform Si at
high-speed by a factor of 20 at the same sensitivity level.

-109-

P Alo.sGao.s-'• 1.5µm
IN

n• GaAa 2.0µm

n• Si substrate

Figure 5.14 Schematic drawing of a ridge-waveguide stripe geometry laser used in the high-speed
modulation experiment.

The growth of the GaAs-on-Si p-i-n detector is the same as that of lasers except
the growth temperature was at 600°C for the entire p-i-n structure. The doping level
for both p and n region is lxl0 18 cm- 3 • The structure of the p-i-n detector is shown
in Figure 5.17. The width of the intrinsic region varies from 1 to 3 µm to balance
the transit time and the RC time constant. The area of the junction is 70x 100µm 2 •
The measurement apparatus is shown in Figure 5.18. The detector was biased through
a microwave bias tee and illuminated with 5 ps optical pulses from a synchronously

-110-

DC Voltage

Pulse
Generator

Source

High Speed
Microwave

p-1-n Photodiode
(Av-10 GHz)

Laser Diode

Bias Tee

hW

• Microwave

...-----4 Bias Tee

Microwave
Sweep
Oscillator

Microwave
Spectrum

Analyzer

Figure 5.15 Schematic drawing of the experimental set-up of high-speed modulation measurement.

pumped mode-locked dye laser (>. =6000 A) at a 100 MHz repetition rate. The output
of the detector was then measured both with a sampling oscilloscope and a microwave
spectrum analyzer. For a 2µm intrinsic region, the impulse response shows a 45 ps
FWHM, corresponding to a 3dB bandwidth of 4 GHz (Figure 5.19). This compares very
well with the results obtained with identical GaAs-on-GaAs p-i-n detectors fabricated
using the same process [18].

§ 5.9 Conclusions

In this chapter, we have reviewed and discussed the details of the popular methods

-111-

+10dB

--

0 dB

~f--Q t-,

>-- re,_ P' _..n --0- ,,0

"O

g -10dB

'\

O>

.Q

-20dB

C\I

-30d8
100MHz

1GHz

2GHz 3GHz

10GHz

Frequency

Figure 5.16 The modulation response versus frequency for a 10 x 380µm 2 stripe under direct
microwave current modulation showing a 3dB bandwidth of 2.5GHz.

used by us and other researchers in the field to grow high quality GaAs-on-Si substrates. Defect reduction is shown to be a major technological barrier in the growth of
high quality GaAs-on-Si. Antiphase domain disorder can be avoided by the use of an As
prelayer, and dislocations can be reduced by use of a tilted substrate. Very satisfactory
results have been obtained using an As prelayer coverage and low substrate temperature at the beginning of the growth. Record low threshold current density lasers have
been demonstrated, which eventually led to the first current-injected room temperature CW operation of GaAs-on-Si lasers. The fabrication of ridge-waveguide geometry
stripe lasers is described. high-speed current modulation of GaAs-on-Si lasers has been

-112-

1iw

1w Wire Bond

Crllw Bond Pad (50µm dia)

p•GaAs(Be)(0.1µm)
Uldoped GaAs (d)

~====~-n+ Ga.As(Si)(~3µm)

n+ Si substrate

Figure 5.17 Schematic drawing of a p-i-n GaAs-on-Si photodiode.

demonstrated, which opens the door to chip-to-chip communication in a Si VLSI system
(19]. High-speed photodetectors have also been demonstrated in this study.

-113-

DC Voltage
Source

Modelocked
Oye Laser

...A_A__

Zo

Microwave

-!-

Bias T

SaJ'll)ling Oscilloscope
or
Microwave Spectrum
Analyzer

r'\./V\.rf> - -

Photodiode

Figure 5.18 Schematic drawing of the experimental set-up used in the measurement of a p-i-n
detector.

§ 5.10 References

[1] R. C. Henderson, J. Electrochem. Soc. , 119, 772(1972).
[2] R. Houdre and H. Marko<;, CRC Critical Review, to be published.
[3] W. I. Wang, Appl. Phys. Lett. , 44, 1149(1984).
[4] R. Kaplan, Surf. Sci. , 93, 145(1980).
[5] R. Fisher, H. Morko<;, D. A. Neumann, H. Zabel, C. Choi, N. Otsuka, M. Longerbone, and L. P. Erickson, J. Appl. Phys. , 60, 1640(1986).
[6] W. T. Masselink, T. Henderson, J. Klem, R. Fisher, P. Pearah, H. Morko<;, M. Hafich,
P. D. Wang, and G. Y. Robinson, Appl. Phys. Lett. , 45, 1309(1984).
[7] see Ref. [2] for more details.
[8] N. Otsuka, C. Choi, L. A. Kolodziejski, R. L. Gunshor, R. Fisher, C. K. Peng,
H. Marko<;, Y. Nakamura, and S. Nagakura, J. Vac. Sci. Technol., B4, 896(1986).
[9] R. Fisher, D. Neumann, II. Zabel, H. Marko<;, C. Choi, and N. Otsuka, Appl. Phys. Lett.,

-114-

"O
CV

rn

--

.......
1,~

a.

rn

CV

c::

\~~

-5

::,
"O

~ ,~ v~

-10

-15

0.1

0.2

0.5

10

FreQuency (GHz)

Figure 5.19 The modulation response versus frequency of a GaAs-on-Si p-i-n photodetector
showing a 3dB band width of 4GHz.

48, 1223(1986).

[10] C. Choi, N. Otsuka, G. Munns, R. Houdre, H. Marko<;, S. L. Zhang, D. Levi, and
M. V. Klein, Appl. Phys. Lett. , to be published.
[11] R. D. Dupuis, J.P. Van der Ziel, R. A. Logan, J.M. Brown, and C. J. Pinzone,
Appl. Phys. Lett, 50, 407(1987).
[12] H. Z. Chen, A. Ghaffari, H. Wang, H. Marko<;, and A. Yariv, Appl. Phys. Lett. ,
51, 1320(1987).

[13] I. Hayashi, M. B. Panish, P. W. Foy, and S. Sumski, Appl. Phys. Lett. , 17,
109(1970).

-115-

[14] H. Z. Chen, A. Ghaffari, H. Wang, H. Morko<;, and A. Yariv, Opt. News, 13, Oct. ,
4(1987), Dec. , 12(1987), also Photonic Spectra, 21, 36(1987), and Semicon. Int. ,
10, 22(1987).

[15] H. Z. Chen, J. Paslaski, A. Ghaffari, H. Wang, H. Morko<;, and A. Yariv, IEEE
IEDM Tech. Digest, 238(1987).

[16] H. Z. Chen, J. Paslaski, A. Yariv, and H. Morko<;, Research and Development, 61,
Jan. , 1988.

[17] K. Lau and A. Yariv, Semiconductor and Semimetals, ed.

W. T. Tsang, (Aca-

demic, Orlando, Fl, 1985), Vol. 22, p. 69.

[18] J. Paslaski, H. Z. Chen, H. Morko<;, and A. Yariv, Appl. Phys. Lett., 52, 1410(1988).
[19] H. Z. Chen, J. Paslaski, A. Yariv, and H. Morko<;, Opt. News, 14, 24(1988).

-116-

Appendix I
Operation and Maintenance of an MBE System

§ 1.1 An introduction
The operation and maintenance of an MBE system has been characterized as a black
art because of the tremendous difficulty in teaching it. No manual can cover everything

that can happen during MBE operations, and one has to make correct judgments and
swift actions. There are however a few things that are established and proven effective
through our extensive use of MBE. The purpose of this chapter is not to present a
general review of MBE operations, but to present the operating procedure we use with
regularity. Although some of the procedures only apply to our Riber-2300 system, most
of them are quite general. The interested reader can use this to supplement the MBE
user's manual [1] and a general book on high vacuum systems [2].
An MBE system contains a UHV ( ultra-high vacuum) processing environment
which employs unparalleled cleanliness, a large heating power input of about lkW,
and frequent mechanical shutter movements. In order to satisfy these demanding specifications, more complex molecular beam and crystal growth monitoring instruments
are being introduced into the system, making it impossible for anyone not familiar with
MBE to fully understand its operation and appreciate its enormous capability. There-

-117-

fore, we begin with a discussion of the minimum requirements of an MBE system.

§ 1.2 Minimum system requirements

In general, complexity implies poor reliability. Such is the case for a fully equipped
MBE system. Each component of an MBE system has a lifetime of say, 3 years, but
there are so many of them operating at the same time. For example, there are: 8
effusion cells and shutters; 1 RHEED system; 1 mass spectrometer; 10 thermal couples;
1 sample manipulator; 3 major viewports; 2 vacuum gauges; 8 power supply units; 1
flux gauge; ... etc. As a result, statistically, we can always expect some kind of problem
with MBE at any given time. Therefore, the most important thing is to know which
components are absolutely essential to a good crystal growth and how to cope with the
failures of non-essential parts.
Let's first look at the important components of an ideal MBE operation and determine which ones are absolutely required.

Vacuum.
No leak, especially on the growth chamber, is allowed.

A leak will introduce

0 2 into the vacuum system which is very detrimental to the growth of high quality
GaAs/AlGaAs layers since it forms a deep trap at mid-bandgap where the impurity
recombination is strongest. A leak also results in a high density of surface defects.
Without LN2 cooling, the system pressure can be as high as 5 x 10- 6 torr, while it
should reach at least 2 x 10- 9 torr or lower with LN2 cooling .

Pyrometer
This is used to monitor the temperature of the substrate surface before and during
a growth. Ideally, a pyrometer can provide valuable information to an operator since
growth temperature is probably the single most important parameter. However, the
viewport in front of the pyrometer slowly gets coated by As, and sometimes almost no

-118-

light can go through it. When this alone happens, the operation of MEE should be
continued. To obtain the substrate growth temperature, which can be used to regulate
the heating power, one needs to go back to the previous record when the viewport
was not coated and find out the pyrometer reading and the substrate heating power
(P

= IV) and use the same power P.

RHEED
RHEED (Reflection High Energy Electron Diffraction) provides dynamic information on the surface structure of the substrate and growing films; in particular, it signals
when the oxide layer on the surface has desorbed and when growth should begin. If the
screen in front of the RHEED is coated by As, one can go back to the previous record
when the RHEED was working and find the heating power at which the oxide film was
thermally removed.

Tantalum shutters
If the Ga or As shutter can not be closed, growth may proceed since we always

need Ga and As. But if either one of them can not be opened then we must repair it.
The Al shutter has to be opened if one needs to grow AlGaAs; a malfunctioning Al
shutter restricts the growth to GaAs alone. Si and Be shutters are needed to be open
at least (if they don't close) if a p-n junction ( such is the case of a laser) is to be grown.
When Si or Be shutter can not be closed, the molecular beam from the cell has to be
cut down by reducing its heating power as rapidly as possible.

Computer
We use a computer to automatically control the heating powers by sending temperature settings to the power supply regulators, recording temperatures, and synchronizing
the cells. If the computer is down, one can simply put the control switch on regulator
to "local" instead of "remote" mode and operate manually.

-119-

Flux gauge
A flux gauge measures the beam flux from a cell. It can tell us for example, the
Ga to As ratio (normally around 1:6). This is not very important since we always try
to find the minimum As flux for the purpose of saving As and to obtain better AlGaAs
growth quality.
The minimum requirements of a working MBE system, therefore, include a good
vacuum and several good shutters and a detailed record of each previous operation.

§ 1.3 The proper vacuum pumping procedure

An MBE system usually has four ( at least three) pumping stages. A first mechanical
pump brings pressure to 100 mmHg. A cryosoption pump then brings it to 10- 4 torr,
followed by a titanium sublimation pump (TSP) which brings it to 10- 5 torr. Finally, an
ion pump pumps it down to 10- 7 --+ 10- 11 torr. The use of TSP is optional but usually
is recommended since it will help the ion pump last longer and the TSP -filaments can
be replaced easily. After the MBE system is opened, it should be flushed a few times
with N 2 (to remove water vapor, etc.) and finally be filled with clean N 2 (N2 has a high
pumping efficiency in cryosorption pumps). At first, the mechanical pump is used to
stream out N 2 • Then three cryosorption pumps should be used in the following way:
the first one is open as long as the streaming-out of gas molecules lasts (because the
N2 molecules can stream out the hard-to-pump argon molecules); the second pump is

open for as long as it pumps; and finally to the third pump. When the third pump
stops pumping, the TSP should be turned on. After a short outgasing of a -filament
inside the TSP ( now the valve is still open to cryosorption pumps), the valve to the
outside is closed and the TSP pumps down fast. When the TSP stops pumping, the
ion pump is switched on to reach a UHV environment.

-120-

MBE system baking
Each time the MBE system has been opened to air for repair and/or recharge, it has
to be baked. The purpose of baking is to heat up the MBE system to a temperature
high enough that most gas molecules condensed during opening will outgas, yet the
temperature is low enough that the ion pump can handle the pumping. The heating
is done with heater tapes rather than with the baking oven provided by Riber. The
heating power can be directly controlled by the electrical current. The ion pump is
equipped with a protection mechanism that will turn it off when the pressure reaches
above lx 10- 4 torr, so a safe baking pressure is slightly below 5x 10- 5 torr. The ion-pump
can also be baked at the same time.
The three cryosorption pumps should be baked and pumped while not being used
to maintain their pumping ability.

§ 1.4 Handling and changing source materials

Whenever a source material ( usually As) runs out, it has to be refilled. Whenever
we open the MBE to replace or repair something, the source materials in the effusion
cells become oxidized and should be replaced. As is often the case, we wait to open the
MBE system to repair some broken parts and replace or refill some source materials at
the same time.
The source materials used in a MBE system are contained in pyrolitic boron nitride
(PBN) crucibles which have a impurity level of less than 10 parts per million, and do
not dissociate below 1400° C. Before we change an old crucible, the new one is always
put in a separate small vacuum system and outgased at 1200° C for several hours in
a Knudsen cell. The source materials should be handled differently. Al chunks can
be lightly etched in HCl and rinsed in deionized water. Ga (purchased with seven 9's
purity) is usually contained in a plastic bottle which needs to be kept at about 40° C in

-121-

liquid form. As chunks are usually concealed in a glass ampule under an inert gas since
As reacts with 0 2 easily. Be is usually contained in a glass bottle. Cleaned Al chunks
can be put into a crucible with a tweezers; Ga liquid can be poured into a crucible; and
As chunks can be dumped into a crucible carefully. Be flakes have to be handled with
extreme care since Be is highly toxic. A plastic glove box is used to handle Be; but
fortunately, Be material can last a very long time and only needs to be refilled about
every 5 years. Si is available in ingot form, and it can be loaded into the MBE system
without further cleaning.
The Al cell, when not in use, should be kept above 700°C since the melting point
of Al is around 550-600 °C, and a rapid heating or cooling through this point causes
a sudden change of volume, which can crack the PBN crucible and damage the entire
effusion cell. The Ga cell is kept at 200° C when not used for the same reason.
The As cracker cell (manufactured by Perkin-Elmer) has two main parts: a large
crucible in the back to hold enough As material that can last for a year, and a long and
thin Mo tube in the front to crack As 4 to As 2 • The Mo tube should always be heated
before the source As gets hot to prevent As from clogging in the Mo tube.
Finally, there is the problem of hole-burning on Ga and Al shutters (unfortunately
the Riber specialists have not been able to explain the cause) which occurs every year.
The solution to this is to keep Ga and Al cell temperatures down as much as possible
before the shutters are opened.

§ 1.5 Calibrations

Growth rate of GaAs and AlGaAs, Al mole fraction, Si and Be doping levels all
change after the MBE system is opened.
performed before any device can be grown.

Therefore, many calibrations have to be

-122-

Doping concentration calibrations
Two measurements are performed: the Hall measurement, and the capacitancevoltage (C-V) measurement.

In the case of Hall measurement, both the mobility and concentration of free carriers
are obtained. The measurement can also be done at liquid nitrogen temperature 77K
in a specially designed dewar to obtain the effect of impurity on carrier mobility. A
sample with a constant doping profile and a given thickness (e.g., lµm) is grown by
MBE for the Hall measurement. To make ohmic contacts for a four-point van der Paul
measurement, four small In balls are pressed onto the four corners of a square sample,
which is then annealed in hydrogen gas for 20 seconds at 400°C. The sample is then
put into a specially designed holder for measurement.

In the case of C-V measurement, there is no need for making any ohmic contact
since the measurement relies on the formation of a Schottky contact between Hg and
sample by a commercially available mercury probe that allows liquid mercury to form a
Schottky barrier to GaAs. The capacitance of this reverse biased diode is then measured
as a function of biasing voltage and information on doping level can then be obtained.

Growth rate calibrations
Since GaAs is grown at 600°C and AlGaAs at 720°C, we need to measure rGaAs(600),
rGaAs(720), rAIAs(720), where r refers to the growth rate, with the help of a scanning

electron microscope (SEM) together with a photoluminescence setup (PL) known as
cathodoluminescence (CL). We need to grow, for example, one hour of each material,
and then put this stack of materials into the CL system to measure the thicknesses as
well as the cathodoluminescence from each layer and calculate the growth rates and Al
mole fractions.
Al mole fraction x in Al,, Ga 1 _,,As is related to the direct energy bandgap in the

-123-

following way

Ef(x)

1.424 + l.247x eV

for x

< 0.45

and

Ef (x)

= 1.424 + l.247x + l.147(x - 0.45) 2 eV

for x

2: 0.45

(3.1)

Therefore, if we know Ef (x) we can determine x.

E(x) can be obtained from the spectrum measured by CL using the following relationship:

E(liw)

= -12398
- ~ eV

(3.2)

Apeak(A)

where >.peak is the peak of PL curve. Thus x can be determined. Once x is determined,
one can easily estimate the growth rate of AlxGa1 _xAs by

rAIAs

= - µm/h
l-x

and

rAIGaAs

raaAs

+ r A/As

(3.3)

and use the result as a double-check of growth rates.

§ 1.6 Conclusions

In this chapter, the techniques of daily maintenance of an MBE system are described and are discussed from an operations point of view. Although there are many
sophisticated instruments installed in an MBE system, only a few of them are required
for growing high quality epitaxial material. The key to a consistent and long-lasting
MBE operation is keeping a good record of all MBE data including substrate temperature, heating power, and vacuum level at all stages for each growth.

-124-

§ I. 7 References

[1] Rib er user's manual.

[2] J. F. O'Hanlon, A User's Guide to Vacuum Technology, John Wiley and sons,
New York, 1980.

-125-

Appendix II
GaAs wafer cleaning procedure

The procedure we have been using is the same one used by the Morkoc,; group at
University of Illinois, and it has the following steps:
1. Boiling substrate in trichloroethylene (TCE) several times. The purpose is to re-

move any wax or oil- based contaminations on a wafer. This step can also be done
with a hot (70°C) H2S0 4 solution with similar effect.
2.

Rinse substrate in acetone, methanol, and deionized water several times.

The

purpose is to remove TCE with acetone, remove acetone with methanol, and remove
methanol with water, since acetone does not dissolve very well in water.
3. Blow dry the substrate with filtered nitrogen. This is important because the next
few steps involve the use of pure H 2 S0 4 •
4.

Put substrate in hot (70° C) H 2 S0 4 for 5 minutes.

The purpose is to remove

any residual organics left from previous cleaning. The temperature of H 2 S0 4 has
to be kept below 100°C, otherwise there will be surface reactions between GaAs
and H 2 S0 4 as we have experienced. When this happens, the process should be
continued since this bad-looking surface layer will be etched away next.
5.

Put substrate in cold (25°C) H 2 S0 4 for 5 minutes, then transfer it to 4:1:1

-126-

(H2S04:H202:H20) to etch for 5 minutes.

The purpose is to etch away a few

microns of GaAs surface material that may contain surface contaminations such as
carbons, structural defects, etc.
6. Transfer the substrate to another cold H 2S0 4 for 1 minute, and rinse in running
deionized water for 5 minutes. The long rinsing removes any H 2S0 4 left on the
surface.
7. Transfer the substrate to HCl for 1 minute. This will remove the very thick oxide
layer on the surface which resulted from a previous 5-minute 4:1:1 etch.
8.

Rinse in deionized water again for 5 minutes, and blow dry with filtered nitrogen.

-127-

Appendix III
Technique of growth interruption

It has become increasingly popular in the MBE community to apply growth interruption techniques to improve interface quality. The purpose of a growth interruption
is to allow the growing surface to relax ( or to reconstruct) following a change in atomic
composition (such as Al grading in AlGaAs). Such a technique can improve the quality of "inverted" GaAs on AlGaAs. The procedure for the (Ga,Al)As system is the
following:
1. close Al shutter

2. drop substrate temperature from 720°C to 600°C
3. grow a thin layer of GaAs ( one monolayer)
4. close all shutters ( Ga, Si, Be, etc.) except As and wait for 1 minute
5. open Ga, Si, and Be shutters and grow one monolayer of GaAs
6. open Al shutter and raise substrate temperature to 720°C

-128-

Appendix IV
Fabrication of broad area lasers

The fabrication procedure of broad area lasers consists of the following steps:
1. First, cleave a small piece of sample, usually smaller than lcmx 1cm. See if the In

on the back is smooth. If it is then one can use the spinner attached to a vacuum
holder without a double-sided masking tape, to spin on a layer of photoresist. If the
In is rough, the tape should be used. Since we will do a lift-off, a thick photoresist
is preferred. AZ-4400 is used for a layer of thick metal layer and 1350J is used for
a layer of thin metal.
2. Then put the sample in a 85°C oven for 30 minutes as a soft-bake to remove moisture
and harden the photoresist so that we can put a mask on it. Some people prefer to
use a infrared lamp to bake photoresist for 5 minutes; this is not a reliable method
and it only saves about 20 minutes.
3. Photolithography is done with a mask. lOOµm wide openings separated by 150µm
wide photoresists are obtained. If the opening is not clear ( residual resist on surface), the oxygen plasma can be used to remove a thin layer of photoresist on the
sample especially in the openings.
4. The sample is slightly etched before it 1s brought into a metalization evaporator

-129-

where a 1000 A of Au (AuZn is used if the p-doping in contact layer is not very
high) is deposited. Lift-off is then done in acetone with the help of a cotton Q-tip.
5. The sample is then mounted upside down with wax to a lapping block. The back
side with In is lapped until the total thickness of wafer is about 5 mils. It is then
cleaved into desirable sizes for threshold current measurement.

-130-

Appendix V
Si wafer cleaning procedure

The details of this cleaning procedure is listed in the following:
l. Degreasing by boiling in trichloroethylene (TCE) 3 minutes, 3 times, rinse in ace-

tone, methanol, deionized water and blow dry with filtered nitrogen. To remove
heavy metal contamination use 60°C 1:1 (HN0 3 :H 2 S0 4 ) for 5 minutes.

3. 2-minute rinse in H 2 0
4. 1:10 (HF:H 2 0) for 20 seconds
5. 2-minute rinse in H 2 0
6. 5:3:3 for 2 minutes
7. 2-minute rinse in H 2 0, 1:10 (HF:H 2 0) for 20 seconds, 2-minute rinse in H 2 0, go to
step 4 and repeat 4 times.
8. 5:3:3 for 5 minutes.
9. rinse in H 2 0 and blow dry with nitrogen.

-131-

Appendix VI
Fabrication of stripe lasers

The procedure described below is the one we have been using in this thesis work.

It has the 10 steps:
1. Photolithographically define 5 µm wide photoresist stripes along (011) facet.

2. Hard bake for at least 2 hours, sometimes the wafer can be left for overnight.

3. Etch in 1:3:40 (H 2 O 2 :H 3 PO 4 :H 2 O) at 1000 A per minute rate until 0.25 µm away
from the GRIN region (waveguide).
4. Grow silicon dioxide film all over the entire surface. The thickness of the oxide film
is not as important as the smoothness.
5. Photolithographically define 1 µm wide photoresist openings centered on top of the
oxide-covered ridges.
6. Hard bake it overnight. Insufficient baking results in rough edges in oxide etch.
7. Etch silicon dioxide in buffered hydroflouric acid (BF HF) according to the thickness
and etching rate of the BF HF.
8. Photolithographically define 100 µm wide photoresist openings centered over the
ridge.
9. Evaporate on 1000 A Au and lift-off in acetone solution.

-132-

10. Lap the back substrate to 4-5 mils and put on 1000 A Au for back contact.