General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright
owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
You may not further distribute the material or use it for any profit-making activity or commercial gain
You may freely distribute the URL identifying the publication in the public portal
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Downloaded from orbit.dtu.dk on: Aug 10, 2022
Semiconductor optical amplifier-based all-optical gates for high-speed optical
processing
Stubkjær, Kristian
Published in:
I E E E Journal on Selected Topics in Quantum Electronics
Link to article, DOI:
10.1109/2944.902198
Publication date:
2000
Document Version
Publisher's PDF, also known as Version of record
Link back to DTU Orbit
Citation (APA):
Stubkjær, K. (2000). Semiconductor optical amplifier-based all-optical gates for high-speed optical processing. I
E E E Journal on Selected Topics in Quantum Electronics, 6(6), 1428-1435.
https://doi.org/10.1109/2944.902198
1428 IEEE JOURNAL ON SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 6, NOVEMBER/DECEMBER 2000
Semiconductor Optical Amplifier-Based All-Optical
Gates for High-Speed Optical Processing
Kristian E. Stubkjaer
Invited Paper
Abstract—Semiconductor optical amplifiers are useful building
blocks for all-optical gates as wavelength converters and OTDM
demultiplexers. This paper reviews the progress from simple
gates using cross-gain modulation and four-wave mixing to the
integrated interferometric gates using cross-phase modulation.
These gates are very efficient for high-speed signal processing and
will open up interesting new areas, such as all-optical regeneration
and high-speed all-optical logic functions.
Index Terms—Add/drop multiplexer, optical gate, optical pro-
cessing, semiconductor optical amplifier, wavelength converter.
I. INTRODUCTION
F
OR YEARS, there has been a desire to realize all-optical
computers using digital optical elements. Clearly, this is
very ambitious since optical elements lack the packing density
of electronic gates because of the much shorter interaction
length of electrons compared to photons. Nevertheless, it is
very realistic to aim at simple optical-signal processing in
telecommunication networks. The requirements are not for
massive processing, but rather the possibility of simple optical
processing at bit rates close to or beyond the bandwidth of
presently available electronics, i.e., 40 Gb/s and above. The
all-optical processing is especially attractive in the high-ca-
pacity core networks where we want to avoid opto-electronic
conversion. The all-optical functions needed in add–drop and
cross-connect fabric are wavelength conversion, add–drop-mul-
tiplexing (wavelength and time), clock recovery, regeneration,
and simple bit-pattern recognition.
For most of these functions, we need simple gates that can be
controlled optically, as shown in Fig. 1. A gate used to modulate
a CW signal or a pulse train can function as a wavelength con-
verter, or part of an optical regenerator, respectively, whereas
gating of an optical input signal can be used for time demulti-
plexing,e.g.Moreover, optical elements that can performsimple
logic operations such as
AND or XOR will be useful for routing
functions for example.
All optical gates are realized by optical nonlinearities in
both glass and semiconductor material and are relying on
mechanisms, such as four-wave mixing, cross gain, cross-phase
and cross-absorption modulation, or combinations of these.
Manuscript received October 9, 2000.
The author is with the Research Center COM, Technical University of Den-
mark, DK-2800 Lyngby, Denmark (e-mail: ks@com.dtu.dk).
Publisher Item Identifier S 1077-260X(00)11552-7.
(a)
(b)
(c)
Fig. 1. Schematic of optically controlled gate used for: (a) Wavelength
conversion by gating of CW light. (b) Regeneration and wavelength conversion
by gating of clock pulses. (c) Demultiplexing–sampling by gating of optical
signal using clock pulses.
Depending on the transfer function of these gates, inverted or
in-phase output signals result (see Fig. 2). Clearly, we need
modules that
1) operate at low optical power levels;
2) are easily adjustable to the system bit rate and to the trans-
mission protocol;
3) are polarization independent; and
4) can be cascaded in several stages.
This paper is primarily devoted to nonlinear elements based
on semiconductor optical amplifiers (SOAs). The history of
SOAs is going back to the beginning of the 1980s, where
the development effort was clearly motivated by the need
for linear amplification in point-to-point systems. Challenges
included the realization of low-facet reflectivities and high
fiber-to-fiber gains by also reducing the coupling losses (see
for example [1] and [2]). With the arrival of the erbium doped
1077–260X/00$10.00 © 2000 IEEE
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 15, 2010 at 04:04 from IEEE Xplore. Restrictions apply.
STUBKJAER: HIGH-SPEED OPTICAL PROCESSING 1429
Fig. 2. Schematicof optically controlled gate and the dependency of the output
signal on the transfer characteristic of the gate.
fiber amplifier (EDFA) around 1990, the SOAs were, however,
almost out-competed as linear amplifiers. The EDFAs featured
lower noise levels and also much better crosstalk properties for
multichannel amplification, due to the much longer recombina-
tion times for the excited states in Er compared to the very fast
carrier dynamics in semiconductor material. SOAs still have a
potential price advantage over EDFAs since they are powered
directly, so they may be useful as amplifiers in access networks.
Moreover, they can provide solutions outside the 1500-nm
band where fiber-based amplifiers are more difficult to realize.
II. XGM G
AT E
The nonlinear behavior that is a drawback for the SOA as
a linear amplifier makes it a good choice for an optically con-
trolled optical gate. First reports of optically controlled SOA-
gates were made in [3] and [4]. In both cases, cross-gain modu-
lation (XGM) was explored: The input signal is used to saturate
the gain and thereby modulate a CW signal (probe) at the de-
sired output wavelength.
The wavelength conversion application has been a very
strong driving force behind the investigations of XGM-gates.
By the middle of the 1990s, it was very clear that wavelength
division multiplexing (WDM) is a very competitive way to
upgrade for higher transmission capacity. Moreover, it became
obvious that WDM networks with all optical cross connects
will be needed for efficient transport of information, e.g.,
[5]–[7]. Wavelength conversion–translation within the network
or at its interfaces is needed for efficient dynamic transport
reconfiguration, high-level restoration, and utilization of the
fiber bandwidth [8], [9]. The converters make it possible to
assign wavelengths on a link-by-link or a subnetwork basis,
thereby relaxing the requirements to the wavelength precision
throughout the whole network [8]. Moreover, wavelength
conversion eases the recovery from link or node failures by
allowing for local rather than global reconfigurations in the
network, e.g., [8]–[14]. Thus, there are very good arguments
for pursuing efficient wavelength conversion.
The cross-gain modulated (XGM) gate is extremely simple
to assemble. It is polarization insensitive because of polariza-
tion-independent SOA gain, and it is very power efficient.It also
turned out that the gate can be extremely fast, and, by 1998, bit
rate capabilities of 100 Gb/s were reported [15], [16]. At first
glance, it seems impossible to reach this speed due to the rela-
tively slow carrier dynamics with lifetimes in the order of 100
psec [17]. Detailed analyzes can, however, explain the signifi-
cant role of gain saturation in achieving high speed [18]. The
prospects of achieving even higher bit rates look fine with the
use of quantum dot material. Pump-probe experiments reveal
very fast gain dynamics [19] in amplifiers made from this ma-
terial.
The XGM gate has a number of shortcomings, such as in-
version of the input–control signal and the relatively large chirp
of the output signal due to the large gain modulation. Never-
theless, the gate has been used with fine results in a number of
switch block experiments, e.g., [20], [21]. Moreover, the gate
has been used to pioneer very interesting work on format con-
version from RZ to NRZ and vice versa [22]. OTDM to WDM
transmultiplexing can also be achieved [23], [24]. The gate has
also been used for demonstration of header erasure and replace-
ment in various optical packet switching schemes [25], [26]. It
remains an interesting challenge to come up with new combina-
tions of such gates to achieve new functionalities.
SOA gates exhibiting optical bistability have also been real-
ized using SOAs or laser diodes with higher facet reflectivities,
e.g., [27], [28]. The bistability is a result of the interaction be-
tween the gain and index variations in the resulting cavity. These
bistable elements do, however, have limited bit-rate capabilities
in the order of 1–5 Gb/s, making them unattractive compared
opto-electronic alternatives.
III. XPM G
ATES
Gates with better performance are achieved by placing SOAs
in interferometric configurations, such as those shown in Fig. 3.
In these gates, the optical input signal controls the phase differ-
ence between the interferometer arms through the relation be-
tween the carrier density and the refractive index in the SOAs
(cross-phase modulation, XPM); thereby a CW light or a pulse
train can be gated [29] or control pulses can be used to gate the
input signal.
For stable operation, the XPM converters must be integrated.
The first monolithic structures reported [30] were based on
Michelson interferometers [Fig. 3(b)], realized by cutting
sections out of the 4
4 space switch with SOAs made by
Ericsson Components [31]. An early realization of a two-port
Mach–Zehnder structure was based on a back-to-back coupling
of the Y-lasers made by Alcatel SEL [32]. Following these early
versions of XPM gates, an impressive activity on monolithic
integration of interferometric gates has taken place [33]–[39],
making these gates one of the test grounds for monolithic
integration of active optical elements.
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 15, 2010 at 04:04 from IEEE Xplore. Restrictions apply.
1430 IEEE JOURNAL ON SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 6, NOVEMBER/DECEMBER 2000
(a)
(b)
Fig. 3. Optical gates using MZI and MI structures with SOAs in the
interferometer arms.
As mentioned, the Mach–Zehnder structures started with
two-port devices, but now they have been developed into
structures with separate waveguides for coupling the input
signal into only one of the interferometer arms, as shown in
Fig. 3(a). The idea is that the input signal depletes the carrier
concentration in only one of the SOAs, thereby creating the
wanted phase difference between the two interferometer arms
in a very efficient way [34]–[37].
The Mach–Zehnder gate has even been refined using bimodal
waveguide structures [39]. They allow the input signal and the
probe signal to co-propagate in the structure using different spa-
tial modes. Structuresincluding integrated CW light sourcesand
structures with integrated optical preamplifiers are also reported
[40], [41].
Michelson interferometric gates have simpler structure since
they offer direct access for the input signal to the SOAs [30],
cf. Fig. 3(b). The MI converter has a reflective facet, making it
a folded version of the MZI converter. Nonlinear loop mirror
configurations with SOAs like the SLALOM (semiconductor
laser amplifier loop mirror) and the TOAD (terahertz optical
asymmetrical demultiplexer) are also reported [38], [42]–[44].
Such gates have the advantage of being inherently balanced, but
only RZ signals can be handled. Another interferometric scheme
that has been used in experiments is the UNI that requires only
a single SOA [60].
Hybrid integration of SOAs on Si-PLCs has been reported
as an alternative to the monolithic integration of the interfero-
metric gates [45]. The approach is interesting since waveguide
technology in Si is relatively mature.
The interferometric converters have the advantage of very
steep transfer functions enabling extinction-ratio regeneration
of the gated signals. Only small input-signal levels are needed to
introduce a
phase difference between the interferometer arms,
so that a very efficient conversion is obtained almost indepen-
dently of wavelength. Because of the small modulation associ-
Fig. 4. Schematic for differential operation of Mazh–Zehnder gate to create
short-switching windows.
ated with the phase shift, the frequency chirp of the output
signal will also be small compared to, e.g., the XGM converter
[30], [46], [47].
Besides signal waveform and spectral reshaping, the inter-
ferometric gates have high optical SNR ratios for the output
signals. Moreover, the noise is redistributed due to the transfer
function [48], [49]. As a result, the noise is accumulating less
rapidly than fora chain ofoptical amplifiers.This allowsfor cas-
cading of several gates [50]. The regenerative properties of the
interferometric gates are important for construction of all-op-
tical cross connects, in which other components-like amplifiers
may degrade the signal quality.
Like the XGM gates, the interferometric XPM gates have
achieved high-speed operation. Thus, wavelength conversion at
100 Gb/s was recently achieved in a Mach–Zehnder structure
[51].
Even faster operation can be achieved with the interfero-
metric structures by controlling both arms of the interferometer
in a differential way, as shown in Fig. 4, thereby creating very
short switching windows (e.g., [52] and [53]) as is demon-
strated experimentally in Fig. 5 [54]. It is clearly seen that the
trailing edges become steep when the differential scheme is
applied. Gates operated in the differential configuration have
successfully been used for time demultiplexing from 160 to
10 Gb/s. Moreover, very elegant add–drop into an OTDM bit
stream can be achieved with a Mach–Zehnder interferometer
[55]. An extension of the add–drop experiment to 80 Gb/s (also
using the Mach–Zehnder from ETH) is shown in Fig. 6 [56].
Recently, the differential-control scheme has also been used to
demonstrate wavelength conversion at an impressive bit rate of
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 15, 2010 at 04:04 from IEEE Xplore. Restrictions apply.
STUBKJAER: HIGH-SPEED OPTICAL PROCESSING 1431
(a)
(b)
Fig. 5. Examples of gated output waveforms (a) without and (b) with
differential control applied [54].
168 Gb/s [57]. High-speed operation has also been achieved,
using nonlinear loop mirrors with SOAs demonstrating demul-
tiplexing from 250 Gb/s [58]. For demultiplexing of OTDM
signals, the semiconductor-based interferometers do not yet
achieve the speed of fiber-based NOLMs (e.g., [59]), but
they have the advantage of being very compact and therefore
potentially simpler to use in real systems.
The first packaged MZI structures have already been realized
and tested in field trials. It should be noted that the gates still
need refinement and as does the schemes for control of their
operating point, but it is interesting to see how fast the devel-
opment of these advanced optical gates has been since the start
around 1993. The interferometric SOA gates have turned out to
be important building blocks for development of all-optical re-
generators and simple signal-processing elements as described
below.
IV. F
OUR-WAVE MIXING IN SOAs
Instead of the cross-gain and cross-phase modulation in
SOAs, it is also possible to utilize four-wave mixing (FWM).
In fact, impressive work on wavelength conversion with FWM
in SOAs was published already in 1989 [61] and numerous
results have been published since then. The FWM scheme (see
Fig. 7) is inherently fast and the gates have the advantage that
many WDM channels can be handled simultaneously [62].
The input-to-output signal efficiency of the gate decreases
with the wavelengthseparation of the pump and the input signal,
but experiments using SOA gates with very long cavities have
resulted in conversion efficiencies approaching 0 dB [63], which
is important for good optical SNR at the output.
Clearly, the output signal wavelength depends on both the
pump (
) and the input signal ( ) wavelengths, so the pump
Fig. 6. Add-and-drop functionality at 80 Gb/s using an SOA based
Mach–Zehnder interferometer. Top trace is the 80 Gb/s input signal (a),
whereas the two lower traces show the 7
2
10 Gb/s signal with a vacated
timeslot (b), and the dropped 10 Gb/s signal (c) [56].
must be tunable even for converters with fixed output wave-
length. Moreover, two pumps will be needed to ensure polar-
ization-insensitive operation since the FWM process is polar-
ization sensitive [62], [64]. Because of the relatively complex
pumping scheme, FWM gates will probably only be used at bit
rates above 100 Gb/s. Wavelength conversion at 100 Gb/s has
been achieved for conversion over 3.2-nm wavelength [65] and
impressive results have also been achieved for conversion over
larger wavelength spans. Examples are 40 Gb/s over 24.6 nm
[66], and 2.5 Gb/s over 80 nm [67].
Also, time-demultiplexing from 100 to 10 Gb/s is reported
with FWM gates [68], and the gates have been used for clock
extraction of a 6.3-GHz clock from a 400-Gb/s signal [69]. An-
other potentially important application is dispersion compensa-
tion by midspan spectral inversion where the optical phase con-
jugation of the FWM process is useful [70]. It should be noted
that very impressive results on midspan spectral inversion have
Authorized licensed use limited to: Danmarks Tekniske Informationscenter. Downloaded on January 15, 2010 at 04:04 from IEEE Xplore. Restrictions apply.
![Fig. 8. Scheme for improved optical regeneration using the Mach–Zehnder interferometer as an invertingAND gate [76].](/figures/fig-8-scheme-for-improved-optical-regeneration-using-the-1o8pjalc.webp)
![Fig. 10. Experimental demonstration of header replacement byXOR gate using a swapping sequence. Measured input and output packet sequences show header (H) and parts of the payload (P) [90].](/figures/fig-10-experimental-demonstration-of-header-replacement-1o7vof6s.webp)
![Fig. 5. Examples of gated output waveforms (a) without and (b) with differential control applied [54].](/figures/fig-5-examples-of-gated-output-waveforms-a-without-and-b-22a8ti1e.webp)
![Fig. 6. Add-and-drop functionality at 80 Gb/s using an SOA based Mach–Zehnder interferometer. Top trace is the 80 Gb/s input signal (a), whereas the two lower traces show the 7 10 Gb/s signal with a vacated timeslot (b), and the dropped 10 Gb/s signal (c) [56].](/figures/fig-6-add-and-drop-functionality-at-80-gb-s-using-an-soa-q6cw47r7.webp)