ICTON 2005
209
Th.B1.1
Mode-Locked Semiconductor Lasers
for Optical Communication Systems
K. Yvind, D. Larsson, L. K. Oxenløwe, J. Mørk, and J. M. Hvam
Research Center COM, NanoDTU, Technical University of Denmark
Ørsteds Plads 345V, DK-2800 Kgs. Lyngby
Tel: (+45) 4525 6366, Fax: (+45) 45936581, e-mail:ky@com.dtu.dk
ABSTRACT
We present investigations on 10 and 40 GHz monolithic mode-locked lasers for applications in optical
communications systems. New all-active lasers with one to three quantum wells have been designed, fabricated
and characterized.
Keywords: mode-locked semiconductor lasers, epitaxial design, noise, optical communication systems.
1. INTRODUCTION
In order for ultra-high speed optical communication systems, using time division multiplexing and optical signal
processing, to be employed in future transmission networks, compact and integratable semiconductor
components need to be developed. The key component to provide the large bandwidth is the mode-locked laser.
In the transmitter, the short-pulse (wide bandwidth) laser can be directly synchronized to a sinusoidal
(narrowband) electrical clock, and it can then provide all the timing functions, such that the requirements on the
electrical components, modulators etc. are strongly relaxed. Later in the network, a low-frequency clock can be
extracted when the laser is synchronized to the bitstream using a suitable clock-recovery scheme, an example of
which will be shown in the end of this paper. Early studies [1] indicated that all-active monolithic semiconductor
lasers would contribute too much excess noise compared to e.g. external cavity lasers and this could limit their
usefulness. However, in this paper we will show that even very simple semiconductor monolithic mode-locked
lasers (MMLLs) are able to deliver very good noise performance.
2. DEVICE DESIGN
The devices investigated are simple ridge waveguide single growth-step Fabry Perot lasers with two electrical
contacts as shown in Fig. 1 (left). The design rules presented below will also be applicable to more complex
lasers with DBR gratings and repetition rate tuning sections as needed for applications in real communication
systems.
In the mode-locked laser, a pulse circulates in the optical cavity with a dynamic broadening in the gain
section and a corresponding pulse shortening in the absorber. The main part of the broadening is due to gain
saturation that also results in nonlinear chirp across the pulse [2]. One way of keeping the pulses short is by
using a strong absorber to absorb the leading part of the pulse. This, however, introduces excess loss which
reduces the efficiency of the laser and, more importantly, increases the noise and instabilities of the laser. A
better approach is to design the gain section such that the pulse broadening is reduced and the requirement for
reshaping is relaxed.
600
n-InP
Active region
(In1-xGa xAsyP1-y)
Ni/Ge/Au
Optical mode
λ=1.55µm
(hν~0.8 eV)
100
0
valence bands
conduction band
0.2
0.4
1
1.2
band positions [eV]
3 QW
1.0
Required
gain 10
GHz
sat,0
[pJ]
2 QW
10
Required
gain 40
GHz
InP:n
-100
0
1 QW
Saturation energy E
BCB polymer
vo
ir
p-InP
200
er
2µ m
300
res
Ti/Pt/Au contacts
ier
,
400
ca
rr
~
m
z
position [nm]
L
m
4)
1(
f=
H
)G
10
(
40
In1-xGaxAsyP1-y (undoped)
InP:p
1, 2 or 3 quantum wells:
width strain
Well: 7.2nm -0.90%
Barrier:2x3.7nm 0.92%
500
1.4
0.1
0
5
10
15
20
-1
Required gain [cm ]
Figure 1. Left/Middle: Schematic of the device with a blowup of the active layers. Right: Calculated gain
saturation energy versus required net gain in the device. The approximate required gain for HR coated lasers of
different frequencies is indicated.
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
The work reported in this paper was supported by the Danish Technical Research Council through the SCOOP
programme and the talent project 26-04-0060.
0-7803-9236-1/05/$20.00 ©2005 IEEE
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 22, 2010 at 07:27 from IEEE Xplore. Restrictions apply.
Th.B1.1
210
ICTON 2005
Table 1. Summary of operating conditions for selected results from 10 and 40 GHz lasers.
Output power includes 3 dB coupling loss.
Number of quantum wells
1
2
1
2
3
4260 / 100
4260 / 55
1070 / 60
1070 / 55
1070 / 50
HR
HR&AR
HR
HR
HR
50, -3.5
50, -3.0
191, -2.7
190, -3.8
240, -4.0
mA, V
9.930, 24
9.973, 24
39.78, 26
39.58, 26
39.53, 26
GHz, dBm
1531
1556
1531
1547
1562
nm
1.8 (0.55)
1.4 (0)
1.1 (0.65)
1.2 (0.65)
1.4 (0.65)
ps (ps/nm)
Time bandwidth product
0.42
0.58
0.38
0.51
0.61
(FWHM)
Fibercoupled output power
0.5
1.6
9.2
8.7
9.0
mW
Device / absorber length
Coating
Gain current, absorber bias
Frequency, RF power
Center wavelength
Pulsewidth (w/ added dispersion)
µm
The epitaxial structures of the active region of the lasers are shown in Fig. 1 (middle). Thick undoped layers are
used to lower the waveguide loss and devices with 1, 2 and 3 quantum wells are fabricated. The absorber facet is
high-reflection coated on all lasers to increase the intensity in absorber.
In Fig. 1 (right), a calculation of the gain saturation energy versus the required net gain is shown. The gain
saturation energy is the pulse energy where the gain is reduced to zero. In practice, only a limited gain saturation
will be tolerated since it results in both pulse broadening and chirp of the emitted pulses. For only a 10%
reduction in gain, the maximum pulse energy should be 1/10 of the value in Fig. 1. For a laser the working point
on the gain curve is fixed by the loss when the laser is operating, which means that the gain saturation energy is
dependent on the number of wells (cmp. Fig. 1 right). For the 40 GHz lasers the required gain is close to the
maximum available using a single well which is the optimum as see in the strongly increasing gain saturation
energy. We would therefore expect that the pulseshaping is more gentle for the 1 QW lasers which would
translate into less chirp, a higher efficiency and less noise. The 10 GHz lasers are four times longer than the
40 GHz lasers and the required gain coefficient is therefore quite low as indicated in Fig 1. We would therefore
expect a lower output power and smaller difference between the different epitaxial structures for these devices.
More details on the influence of the epitaxial design can be found in [3, 4]
3. RESULTS FOR 10 GHz LASERS
The lasers are mounted with the absorber section bonded to a coplanar electrical waveguide, which allows
modulation to be applied for the synchronization to an electrical clock. Operating conditions and pulse
properties for typical performance close to the optimum operation point is shown in table 1. The pulse
performance of the 10 GHz lasers is actually better for a high- and low (5%) reflection coated 2 QW laser (see
Fig. 2 left) than for the 1 QW laser. This is because the difference between the gain saturation energies is not
especially large so that the more efficient modulator in the 2 QW device gives shorter pulses. In Fig. 2 (right)
the absolute phase noise is shown, measured with a photodiode/spectrum analyzer combination, detecting the
pulse train directly at 10 GHz. At low offset frequencies the laser follows the electrical clock (synthesizer)
completely as would be expected, while the laser noise is visible at offsets > 1MHz before the noise floor is
reached at 200 – 300 MHz. The noise for the 2 QW device is higher than that for a 1 QW device due to the
Autocorrelation
(TBP 0.58)
2
Sech (t)
(FWHM 1.4 ps)
-90
0.8
Intensity [AU]
SHG intensity [AU]
0.8
-80
Measurement
2
Sech (f)
(FWHM: 1.8 nm)
1.0
0.6
0.6
FWHM: 3.3 nm
0.2
0.0
1545
0.4
1550
2 QW
(HR & AR)
-100
0.4
1555
1560
Wavelength [nm]
0.2
L(f) [dBc (1 Hz)]
1.0
-110
-120
-130
1 QW (HR)
-140
-150
HP8673C Synthesizer
-160
0.0
-6
-4
-2
0
2
Delay [ps]
4
6
8
10
1k
10k
100k
1M
10M
100M
1G
Offset [Hz]
Figure 2. Left: Autocorrelation of a 2 QW hybridly mode-locked laser with the optical spectrum inserted.
Right: Absolute phasenoise measurement for 1 and 2 QW 10 GHz lasers.
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 22, 2010 at 07:27 from IEEE Xplore. Restrictions apply.
ICTON 2005
211
Th.B1.1
Vabs
50-GHz
photodiode
Igain
laser
microwave
delay
High power
RF amp
-100
Q-band
LNA
RF
Q-band
mixer
IF
LO
V
RF
Spectrum
Analyzer
2QWs 3QWs
white noise
-110
L(f) [dBc (1Hz)]
40-GHz
synthesizer
LNA
25dB/dec
-120
1QW
-130
-140
-150
synt/noise floor
-160
Bias-T
-170
Offset frequency [Hz]
160
1
1.0
0.1
jitter (lower limit-20 GHz) [fs]
0.8
Intensity [a.u.]
ACF Intensity [au]
Measurement
2
Sech (t) fit
∆λ= 2.2 nm
0.6
0.4
0.2
0.8 ps
0.0
1520
1525
1530
1535
Wavelength [nm]
0.01
1QW
2QWs
3QWs
140
120
100
80
60
40
20
0
1E-3
-8
-6
-4
-2
0
2
4
delay [ps]
6
8
10
12
10k
100k
1M
10M
100M
1G
10G
lower integration limit [Hz]
Figure 3. Top left: Residual phase noise measurement setup. Right: Residual phase noise measurement
result and integrated jitter from a variable lower limit to the Nyquist frequency. Lower left: Autocorrelation
of low noise pulse from 1 QW device at “optimal” compression (0.9 ps/nm). Inset: optical spectrum showing
excess spectral components on the high energy side.
higher spontaneous emission noise and stronger pulseshaping for the 2 QW device. If the gain current is
increased, the jitter is lowered slightly, but the pulse is broadened and more chirped which is consistent with the
low gain saturation energy of the gain medium at the low required gain.
For demultiplexing and optical signal processing it is the phase difference between the data signals and the
recovered clock that is of importance. For these applications one would use a wide-bandwidth clock-recovery
(CR) circuit to capture as much as possible of the slow jitter and the remaining high-frequency noise is what is
important. This “uncorrelated” jitter should be less that 1/12 (1/14.2) of the bit period for a 10-9 (10-12) bit error
rate [5]. For the data in Fig 2, the integrated jitter in the range 1 MHz – 5 GHz is ~195 fs for the 1 QW device
and ~300 fs for the 2 QW device but this includes a large contribution from the noise floor of the measurement.
4. RESULTS FOR 40 GHz LASERS
For the 40 GHz lasers the length of the device is only 1 mm and the required gain including the absorber loss is
close to the maximum available for a 1-QW device as seen in Fig. 1 (right). This indicates that it will be possible
to operate it at higher powers, which will ensure low jitter operation while keeping short pulses.
If absolute phase noise measurements are used to evaluate the noise of the 40 GHz lasers the effective noise
floor of a 50 GHz electrical spectrum analyzer (~-125 dBc (1 Hz) for our HP8565E) will mask the contribution
from the laser completely [6]. Residual phase noise measurements (Fig. 3 top, left), where external mixing of the
detected pulse train with the reference clock is used before the spectrum analyzer, is a way for solving the
measurement problem [1, 7]. Since this measurement is relative to the synthesizer the contribution from the
latter at long timescales will not appear because the laser tracks this part of the jitter as shown for the 10 GHz
device. This is an advantage since the noise we are interested in for practical purposes is the difference between
the clock and laser as noted earlier. More importantly, because the carrier has been removed in the external
mixing we can amplify the noise bands and send them into the spectrum analyzer without saturating its mixer.
We are therefore able to follow the fall-off of the side-band noise several decades as shown in Fig. 3 (top right).
Also, when we evaluate the rms jitter we can now integrate all the way to the Nyquist frequency (20 GHz) and
thus include the pulse to pulse jitter, which will introduce unavoidable penalties in the optical signal processing.
As shown in Fig. 3 (lower right), a value of less than 60 fs for the 1 QW device is seen (integrated in the full
measurement bandwidth) which is sufficient for Tbit/s transmission systems. The lack of intracavity filtering
limits the spectral quality of the lasers as shown in the inset of Fig. 3 (lower left) and give rises to the shoulders
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 22, 2010 at 07:27 from IEEE Xplore. Restrictions apply.
Th.B1.1
212
ICTON 2005
on the autocorrelation for compressed pulses. It should be noted that while the pulse width and spectrum can be
modified fairly easy outside the laser cavity, this is not the case for the jitter.
5. CLOCK RECOVERY USING MONOLITHIC MODELOCKED LASER
As an example of an application for the lasers, a 10 GHz clock-recovery (CR) circuit based on a phase-locked
loop using four-wave mixing (FWM) in a semiconductor optical amplifier (SOA) is shown in Fig. 4. This prescaled CR setup, extracts a 10 GHz clock from a high bit-rate pulse train, and uses only semiconductor
components meaning that it can be integrated on a single chip.
A 40-320 Gbit/s optical time division multiplexed sequence is launched into a SOA along with an amplified and
filtered 10 GHz pulse train from a MMLL. The FWM product is extracted and fed to a feedback loop which
controls an oscillator which drives the MMLL. In Fig. 4 (right) the error signal is shown when the feedback loop
is open and the control pulses scans across the data pulses. A clear signal, especially for the lower repetition
rates is seen, which can be used to lock the clock. Due to the filtering and fiber in the loop the laser pulse is
almost 4 ps which is not enough to resolve the 320 Gbit/s, but successful locking to up to 160 Gbit/s has been
achieved [8].
data
MOD
transmitter
clock
electrical
clock
FWM
40-320 Gb/s
OTDM data
1548 1554 1560
OTDM data
10 GHz
SOA
SOA
MMLL
VCO
λFWM
PD
PD
+
-
PI
feedback loop
320 Gb/s
160 Gb/s
80 Gb/s
40 Gb/s
0,10
power [a.u.]
MUX
ML-FRL
0,08
MMLL
FWHM:
3.9 ps
0,06
0,04
0
300
600
time [a.u.]
900
Figure 4. Left: High- speed clock recovery set-up. Right: Error signal at 40, 80, 160 and 320 Gb/s.
Inset: Clock pulse shape.
6. CONCLUSIONS
We have shown that the use of few quantum wells in monolithic mode-locked lasers enables both low noise and
increased output power for 10 and 40 GHz lasers. Such lasers can fulfil the requirements for terabit optical time
division multiplexed communication systems.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
D.J. Derickson, P.A. Morton, and J.E. Bowers: Comparison of timing jitter in external and monolithic
cavity mode-locked semiconductor lasers, Appl. Phys. Lett., vol. 59, pp. 3372–3374, 1991.
S. Bischoff et al.: Pulse-shaping mechanism in colliding-pulse mode-locked laser diodes, Appl. Phys. Lett.
vol. 67, pp. 3877–3879, 1995.
K. Yvind et al.: Design and evaluation of mode-locked semiconductor lasers for low noise and high
stability, OptoIreland Proceedings of SPIE, Paper 5825A-4, 2005.
K.A. Williams, M.G. Thompson and I.H. White:Long-wavelength monolithic mode-locked diode lasers,
New Journal of Physics 6, no. 1, 179 (2004).
K. Yvind et al.: Low-Jitter and High-Power 40 GHz All-Active Mode-Locked Lasers, Phot. Tech. Lett.,
vol. 16. pp. 975-977, 2004.
M. Jinno: Effects of crosstalk and timing jitter on all-optical timedivision demultiplexing using a nonlinear
fiber Sagnac interferometer switch, IEEE J. QE, vol. 30, pp. 2842–2853, 1994.
L.A. Jiang et al.: Quantum-limited noise performance of a mode-locked laser diode, Optics Letters, vol.
27, pp. 49-51, 2002.
L.K. Oxenløwe, et al.: Pre-scaled clock recovery with compact semiconductor devices for ultra high-speed
OTDM systems, in Proc ECOC 2004, paper We3.5.2, Stockholm, Sweden, 2004.
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 22, 2010 at 07:27 from IEEE Xplore. Restrictions apply.