of 3
Electrically pumped hybrid evanescent
Si/ InGaAsP lasers
Xiankai Sun, Avi Zadok,
*
Michael J. Shearn, Kenneth A. Diest, Alireza Ghaffari, Harry A. Atwater,
Axel Scherer, and Amnon Yariv
Department of Applied Physics, MC 128-95, California Institute of Technology, Pasadena, California 91125, USA
*
Corresponding author: avizadok@caltech.edu
Received January 22, 2009; revised March 16, 2009; accepted March 26, 2009;
posted March 30, 2009 (Doc. ID 106493); published April 21, 2009
Hybrid Si/III–V, Fabry–Perot evanescent lasers are demonstrated, utilizing InGaAsP as the III–V gain ma-
terial for the first time to our knowledge. The lasing threshold current of
300-

m
-long devices was as low as
24 mA
, with a maximal single facet output power of
4.2 mW
at
15 ° C
. Longer devices achieved a maximal
single facet output power as high as
12.7 mW
, a single facet slope efficiency of 8.4%, and a lasing threshold
current density of
1kA/cm
2
. Continuous wave laser operation was obtained up to
45 ° C
. The threshold cur-
rent density, output power, and efficiency obtained improve upon those of previously reported devices having
a similar geometry. Facet images indicate that the output light is largely confined to the Si waveguide.
© 2009 Optical Society of America
OCIS codes:
250.5960, 250.5300
.
A monolithic integration of lasers together with Si in-
tegrated electronic circuits has been highly sought af-
ter for decades. Unfortunately, Si is a poor converter
of electricity to light, due to its indirect semiconduc-
tor bandgap. In addition, the epitaxial growth of
standard GaAs and InP based direct bandgap mate-
rials on Si substrates has proved to be a major ob-
stacle, due to the mismatch in lattice constants and
in thermal expansion coefficients [
1
]. Despite these
difficulties, recent years witnessed a reawakening of
the interest in Si-integrated lasers, pursued prima-
rily to support high-rate interchip and intrachip com-
munication within multiprocessor computers [
2
].
Numerous avenues have been followed towards ob-
taining Si-integrated lasers, including Raman ampli-
fication [
3
], rare-earth doping [
4
], and nanocrystal-
line Si structures [
5
]. One particularly successful
approach is based on wafer bonding of AlGaInAs ma-
terial on top of a prepatterned Si-on-insulator (SOI)
wafer [
6
8
]. The bonded structure is designed to sup-
port a joint optical mode, whose profile overlaps both
materials. Using this technology, the AlGaInAs lay-
ers, which include multiple quantum wells, could be
patterned postbonding to produce Fabry–Perot (FP)
[
6
], racetrack [
7
], and distributed feedback lasers [
8
],
the outputs of which are predominantly emitted from
the underlying Si waveguides. The modal gain is ob-
tained by the evanescent tail of the joint mode, which
penetrates into the multiple quantum wells. The
devices are referred to as evanescent hybrid
Si / AlGaInAs lasers [
6
]. A similar approach is imple-
mented in Si-coupled microdisk lasers [
9
], although
their output power is limited to tens of

W.
The AlGaInAs material system is advantageous in
uncooled laser operation at high temperatures, due to
the large conduction band offset [
10
]. On the other
hand, high-quality AlGaInAs layers are relatively
more difficult to obtain, and the reliability of Al-
containing lasers remains a concern [
10
,
11
]. In addi-
tion, Al alloys are prone to nonradiative surface re-
combination, which may elevate the lasing threshold
current. In this work, we demonstrate hybrid Si/
III–V, evanescent FP lasers based on a different
III–V material system, i.e., InGaAsP quaternary
compounds. We find that the threshold current den-
sity and the threshold voltage of the fabricated de-
vices are 30–40% lower than those of the correspond-
ing, previously reported FP devices [
6
]. As one of the
key challenges facing hybrid Si/III–V active devices
is the extent of heat generation [
12
], the reduction in
threshold current and voltage may prove significant.
At the same time, the output power and differential
slope efficiency observed are higher than previously
reported.
The hybrid Si/III–V structure consists of an SOI
wafer and an InGaAsP wafer that were bonded to-
gether. The thicknesses of the buried SiO
2
layer and
the undoped Si device layer were 2.0

m and 0.9

m,
respectively. A waveguide was defined in the Si de-
vice layer using electron beam lithography and sub-
sequent SF
6
/C
4
F
8
plasma reactive ion etching. The
waveguide width ranged between 0.9

m and
1.3

m. The Si to the two sides of the waveguide was
entirely etched, down to the SiO
2
layer. After etching,
the SOI wafer was cleaned by solvents and a 3:1
H
2
SO
4
:H
2
O
2
mixture (10 min at 170 ° C). The details
of the key structural layers of the InGaAsP wafer,
grown on top of a 350-

m-thick InP substrate, are
provided in Table
1
. The wafer design guidelines
closely follow those of [
6
], albeit in a different mate-
rial system.
The bonding procedure began with solvent clean-
ing of both surfaces. A 10-nm-thick oxide layer was
grown on top of the patterned SOI wafer to enhance
the bonding strength. The surfaces of the wafers
were then activated through exposure to oxygen
plasma and bonded together under a pressure of
0.1 MPa at 350 ° C for 1 h. Low temperature, plasma-
assisted bonding was shown to be a powerful tool for
integrating dissimilar material systems [
6
,
13
].
Following the bonding, the InP substrate was re-
moved by HCl wet etching. An 80-

m-wide mesa
structure was formed in the InGaAsP layers, cen-
May 1, 2009 / Vol. 34, No. 9 / OPTICS LETTERS
1345
0146-9592/09/091345-3/$15.00
© 2009 Optical Society of America
tered above the Si waveguide, through photolithogra-
phy and subsequent three-phase wet etching, down
to the
n
-InP contact layer (see Table
1
). The etching
solutions were (a) 1:1:10 mixture of H
2
SO
4
:H
2
O
2
:
H
2
O(
p
-InGaAs layer, 60 s), (b) 2:1 mixture of
HCl: H
2
O(
p
-InP layer, 30 s), and (c) 1:1:10 mixture of
H
2
SO
4
:H
2
O
2
:H
2
O (quaternary layers, 4 min). Metal
contacts were deposited for the
p
side

Cr / AuZn / Au

on top of the remaining
p
-InGaAs layer, and for the
n
-side

Cr / AuGe / Au

on the exposed
n
-InP layer to
the two sides of the mesa. The current flow was lat-
erally confined to a 5-

m-wide channel by means of
proton implantation on its two sides [
14
]. The im-
plantation dosage and proton energy were 5

10
14
cm
−2
and 170 keV, respectively. Finally, the Si sub-
strate was lapped down to a thickness of 50

m, and
device bars were cleaved and annealed at 410 ° C for
10 s. The annealing assists in the diffusion of Zn
from the
p
-side metal contact into the
p
-side layers,
and therefore reduces the resistance of that region.
Figure
1
shows a top view optical microscope image
and scanning electron microscope (SEM) images of
the device’s cross section.
Figure
2(a)
shows the output power and device
voltage versus current (L–I–V curve) of a 960-

m-
long device, mounted on a thermoelectric cooler at
15 ° C. The turn-on voltage was 0.8 V, and the lasing
threshold voltage
V
th
was 1.3 V. The threshold cur-
rent
I
th
was 60 mA, corresponding to a threshold cur-
rent density
J
th
of 1.25 kA / cm
2
. The maximum power
output
P
max
from a single facet was 12.5 mW, and the
differential slope efficiency

diff
for a single facet was
8.4%. The series resistance of the laser was 8

. The
inset of Fig.
2(a)
shows
I
th
as a function of tempera-
ture. Continuous wave lasing was achieved at tem-
peratures up to 45 ° C, and the characteristic tem-
perature of the device was found to be 39 ° K. Figure
2(b)
shows the laser spectrum, whose central wave-
length was 1490 nm. The modal loss

i
was esti-
mated as 28 cm
−1
, using Hakki–Paoli measurements
below the bandgap [
15
]. The measured values of

diff
and

i
correspond to an internal quantum efficiency
of 0.54.
J
th
of 1 – 1.5 kA / cm
2
were obtained for nu-
merous devices, having lengths ranging between
300 and 1500

m.
I
th
of the 300-

m-long devices was
24 mA at 15 ° C, with
P
max
of 4.2 mW.
J
th
and
V
th
of
the devices are about 35% lower than those of previ-
ously reported FP hybrid Si / AlGaInAs lasers [
6
]. At
the same time, the devices’
P
max
is 70% higher, and
their

diff
is 30% higher. At this stage, we cannot yet
Fig. 1. (Color online) (a) Top view of a fabricated device.
(b) SEM overview of a cross section of the device. (c) SEM
close-up view of the device cross section at the Si wave-
guide region. Approximate ion implanted regions are super-
imposed on the image for illustration.
Table 1. InGaAsP Wafer Epilayer Structure
Layer
Material
Thickness (nm)
Bandgap (eV)
Doping

cm
−3

p
-side contact layer
p
-In
0.53
Ga
0.47
As
200
0.77
p

10
19
Upper cladding layer
p
-InP
1500
1.34
p
=10
18
5

10
17
Separate confinement layers
InGaAsP
40
1.08
undoped
InGaAsP
40
0.99
undoped
Quantum wells (1% compressive strain)
InGaAsP


5

7
0.83
undoped
Barriers (0.3% tensile strain)
InGaAsP


4

10
0.99
undoped
Separate confinement layers
InGaAsP
40
0.99
undoped
InGaAsP
40
1.08
undoped
n
-side contact layer
n
-InP
110
1.34
n
=10
18
Superlattice
n
-InGaAsP


2

7.5
1.13
n
=10
18
n
-InP


2

7.5
1.34
n
=10
18
Bonding layer
n
-InP
10
1.34
n
=10
18
1346
OPTICS LETTERS / Vol. 34, No. 9 / May 1, 2009