Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO.

I, JANUARY 1992 93

Chaos in Semiconductor Lasers with Optical


Feedback: Theory and Experiment
Jesper Mark, Bjame Tromborg, and Jannik Mark

Abstract-We present a detailedtheoreticaland experimental The system also deserves attention from the basic
investigation of the nonlinear dynamicsof a semiconductorlaser standpoint, since it represents an important example of an
with optical feedback. The results show that the coherence col-
lapsed state is a chaotic attractor and that chaos is reached for active type of delayed feedback optical system with a be-
increasing feedback level through a quasi-periodicroute inter- havior quite different from that of the well-studied passive
rupted by frequency locking. Furthermore, the coexistence of Ikeda system [171, [ 181. The phase space associated with
two attractors associated with the same external cavity mode, a delayed feedback system is of infinite dimensions, and
but having different relaxation oscillation frequencies, is dem- the study of such systems is important in order to inves-
onstrated and explained.
tigate, to what extent the results previously obtained for
low-dimensional systems carry over to infinite-dimen-
sional systems [ 191.
I. INTRODUCTION Experimental characterization of the behavior of a

S EMICONDUCTOR laser diodes exposed to optical


feedback have been found to undergo a dramatic spec-
tral broadening for a large range of feedback levels [1]-
semiconductor laser diode with optical feedback has
shown that five regimes (I-V) of qualitatively different
effects can be distinguished for increasing feedback level
[5]. The phenomenon is denoted coherence collapse [2] [3]. For very small feedback levels, regime I, only one
and has been interpreted as a manifestation of chaotic dy- external cavity mode exists. In regime I1 the laser shows
namics [2], [4], [6]-[ll]. This paper presents a detailed noise-induced hopping between several external cavity
theoretical and experimental investigation of the nonlin- modes. The rate of mode hopping decreases for increas-
ear dynamics involved in coherence collapse. Brief ac- ing feedback level, and in regime I11 the laser is observed
counts of some of the results presented in this paper were to operate only in one dominant mode, which is the mode
published in [ 121 and [ 131. with minimum linewidth. Coherence collapse occurs in
For applications of laser diodes requiring narrow line- regime IV which extends over several decades of feed-
width operation (e.g., coherent communication systems) back level. It is bounded from above by regime V of sin-
coherence collapse places severe demands on the optical gle-mode narrow linewidth operation, but in order to reach
isolation of the laser diode. Thus, spurious back reflec- this regime of very strong feedback it is necessary to
tions on the order of lop4(i.e., -40 dB) of the emitted antireflection coat the laser facet facing the external cav-
laser beam from various interfaces (e.g., optical fibers) ity or to use a very low bias current [20], [ 111.
are of sufficient magnitude to drive the laser into coher- Theoretical investigations of optical feedback effects are
ence collapse, with the laser linewidth broadening from a usually based on the Lang and Kobayashi rate equations
few MHz to several GHz. On the other hand, intentional [21] which have proven to contain all the dominant effects
optical feedback may be employed to significantly reduce observed experimentally. For weak feedback, regimes I-
the laser linewidth [ 11, [ 141, [ 151 (subhertz linewidth op- 11, a small-signal analysis has been demonstrated by many
eration has been demonstrated using Rayleigh backscat- authors to give a correct description of linewidth, spectral
tering from a fiber [16]) and such external cavity lasers behavior and modulation properties, etc. [22]. For mod-
are commonly used for applications where the spectral erate or strong feedback, regimes 11-V, the nonlinearities
purity of monolithic laser diodes does not suffice. How- must be taken into account. Computer simulations have
ever, the utilization of feedback effects requires either shown that the occurrence of mode hopping [23], [24],
antireflection coated Fabry-Perot lasers or very careful low-frequency intensity fluctuations [25], [7], and the
control of the amount of feedback, in order to avoid the onset of coherence collapse [4], [23], [26] is correctly
regime of coherence collapse. From a practical point of predicted by the Lang and Kobayashi equations. The in-
view it is therefore important to have a detailed under- finite-dimensional nature of the system makes analytical
standing of the behavior of a laser diode with optical feed- investigations very difficult and in order to gain insight
back. into the physics behind the various phenomena, it is there-
fore desirable to employ some kind of simplifying ap-
Manuscript received February 5 , 1991. proximations that reduce the dimensionality of the sys-
The authors are with TFL Telecommunications Research Laboratory,
DK-2970 H$rsholm, Denmark. tem. Several approaches have been taken.
IEEE Log Number 9103724. For feedback levels below about -45 dB one may as-
0018-9197/92$03.00 0 1992 IEEE
94 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO. 1, JANUARY 1992

sume the light intensity to be constant which leaves the Time domain studies are hindered by the very fast time
phase of the electric field as an independent variable of scale (2.0.25 ns) on which the dynamics take place.
the system. This approximation leads to a potential model However, we show that the use of a heterodyne down-
[24], [27], [28] which explains why external cavity lasers conversion technique allows the recording of experimen-
prefer to oscillate at the mode with minimum linewidth tal phase portraits.
instead of the mode with minimum threshold gain. The The system determinant for operation in a specific ex-
model also predicts the experimentally observed rates of ternal cavity mode has several resonances, which are dis-
mode hopping to a surprising accuracy [27]. However, it placed in frequency by approximately the inverse of the
does not apply to the regime of coherence collapse. In the external cavity round-trip time. The resonances closest to
work by Lenstra, Cohen, and collaborators [2], [29], [30], the relaxation oscillation frequency of the solitary laser
the light injected from the external cavity is considered as (i.e., the laser without feedback) exhibit the smallest
a noise source, when the system is in a coherence col- damping. As mentioned above, the route to chaos is ini-
lapsed state. By self-consistency calculations they obtain tiated when one of these resonances becomes undamped.
a statistical description of the collapsed state (see also However, for some choices of parameters, two reso-
[31]) that agrees well with experiments, although some nances may become undamped almost simultaneously.
discrepancies between theory and measurements have re- Our numerical calculations show that in such cases two
cently been detected [30]. A similar approach was used coexisting attractors are created which belong to the same
by Wang and Petermann [32] to put an upper limit on the external cavity mode. Mathematically, the second attrac-
relative intensity noise (RIN) due to optical feedback. The tor is created, somewhat unexpectedly, by a Hopf bifur-
approach cannot, though, resolve the question as to which cation from an unstable external cavity mode. Experi-
route to chaos the laser undergoes. Another way to sim- mental observations of intensity noise spectra with two
plify the system is to use the so-called injection locking strong relaxation oscillation peaks are well known, and
model which was first introduced in [34] and by which were previously explained by noise excitation of the res-
the feedback system is replaced by a laser diode exposed onances of the system. However, our time and frequency
to injection of the stationary feedback field. In [7] this resolved studies of the system clearly demonstrate that the
was shown to explain the characteristic pattern of inten- laser randomly jumps between two periodic solutions,
sity dropouts observed in the time evolution of the inten- which have different oscillation frequency but are asso-
sity for low bias currents [25], [33], [8], [lo]. In [ l l ] we ciated with the same external cavity mode. This is in good
investigated the nonlinear dynamics and the regimes of agreement with our theoretical prediction.
bistability in the injection locking model, and we showed The transition to chaos from the regime V to regime IV
how this can be used to roughly predict the regime of co- for fixed moderately strong feedback and increasing bias
herence collapse for the feedback system. current has been suggested to follow an intermittency
In reducing the dimensionality of a feedback system, route to chaos [8] , [ 101. The time evolution of the inten-
one should, in general, be aware that this may profoundly sity for bias currents around the threshold of the solitary
change the nonlinear dynamics of the system. This was laser is characterized by randomly occurring sudden drops
explicitly shown in [ 181 for the case of the Ikeda system. in intensity followed by a stepwise intensity buildup and
In this paper we study numerically the dynamics of the damped low-frequency relaxation oscillations. Inbetween
full set of rate equations. The results show, however, that the dropouts there may be bursts of noise at the external
in most cases the laser dynamics is confined to low-di- cavity round-trip frequency (see, e.g., [lo], Fig. 3, and
mensional attractors, which may then be characterized [8], Fig. 2). The sudden drops in intensity and subsequent
using well-known techniques from nonlinear dynamics stepwise buildup was, in [7], [ l l ] , explained by the in-
1351. jection locking model as being due to bistable switching
We will mainly focus our attention on the phenomena and dynamic formation of bistability. In [33], [7], [ l l ] ,
(bifurcations) involved in the transition between regime [lo] the peak in the intensity noise spectrum at low fre-
I11 of single mode narrow linewidth operation and regime quencies (0-50 MHz) was associated with the average
IV of coherence collapse, which takes place for a feed- frequency of the power dropouts, while in other publica-
back level in the range of -45 to -30 dB. Our calculated tions [36]-[40] the peak was related to the low-frequency
bifurcation diagrams demonstrate that for increasing feed- relaxation oscillation resonance of the feedback system.
back level the laser undergoes a quasi-periodic (Ruelle- This issue has recently been settled by Sacher and Elsas-
Takens) route to chaos [35] which may be interrupted by ser [41] who have shown experimentally that both contri-
frequency locking. The calculated attractors within the re- butions can be present simultaneously and have measured
gime of coherence collapse have the typical multisheeted the bias current dependence of the frequencies of the two
appearance of chaotic attractors [35]. A small-signal sta- contributions. The relative importance of the contribu-
bility analysis shows that the quasi-periodic route to chaos tions depend, e.g., on the external cavity length, the bias
is initiated by a Hopf bifurcation which corresponds to current, and the laser material. We will not discuss the
undamping of the relaxation oscillations. Experimentally, transition between regimes V and IV any futher in this
we verify this route to chaos through study of the evolu- paper, but there is clearly a need for a more detailed anal-
tion of the intensity noise spectrum with feedback level. ysis.
M$RK et al.: CHAOS IN SEMICONDUCTOR LASERS 95

The occurrence of subharmonic bifurcations has been


reported for an external cavity laser system [42], [39],
[40], but these results relied on a deliberate tilting of the rl r2 r3
external mirror whereby subharmonics of the external Fig. 1. External cavity laser configuration.
cavity round-trip frequency are introduced. Quasi-peri-
odic and frequency locking phenomena, similar to those
reported here, have been observed in semiconductor la-
d
sers which were self-pulsing [43] or coupled to an exter- - N(t) = J - N(t>
- - G(N, E:) E:(t)
nal cavity [44], [45]. However, these results relied on a dt 7s (3)
direct modulation of the bias current pumping the laser
diode, whereby an external frequency is imposed on the where
system. In contrast the laser system investigated here is
A(t) = 40) - 40 - 7) (4)
pumped with a constant (CW) current, and the frequen-
cies involved in the dynamics thus arise internally in the is the phase delay during the external cavity round-trip
laser. time 7 . The instantaneous optical angular frequency w(t)
The paper is organized as follows. In Section I1 we in- +
is given by w = wo d + / d t . Hence
troduce the rate equations and discuss the stationary
solutions (external cavity modes) and their small-signal
stability properties. Section I11 presents numerical bifur- A7 I-7
w(t') dt' = wo + A/7 (5)
cation diagrams for increasing feedback level and anal-
yses of the various attractors encountered. Two examples which shows that A / r is an average frequency deviation
are considered. The first example demonstrates the basic from the oscillation frequency of the solitary laser. The
quasi-periodic route to chaos and the second illustrates the field amplitude Eo(t) is normalized such that Z(t) =
existence of a three-frequency quasi-periodic attractor and VCEo(t)2is the total photon number in the laser wave-
the coexistence of two attractors associated with the same guide, with Vc being the volume of the active region. The
external cavity mode. Section IV presents the experimen- modal gain per unit time is
tal demonstration of a quasi-periodic route to chaos and
the occurrence of quantum noise induced jumps between G(N, E;) = GN(N - No) (1 - d'E;) (6)
two attractors oscillating at the same optical frequency. where GN is the modal gain coefficient, No is the carrier
Finally, Section V summarizes our conclusions. density at transparency, E is the nonlinear gain compres-
sion coefficient, and r is the confinement factor. The in-
11. THEORETICAL MODELAND STABILITY ANALYSIS clusion of nonlinear gain is known to be important for
Rate Equations correct modeling of the dynamics of semiconductor la-
sers. In (2) we have neglected the dependence of the line-
The external cavity configuration where a laser diode is
width enhancement factor a on the photon density, which
exposed to optical feedback from a flat external mirror is
is not yet known. Nsoris the threshold carrier density for
shown in Fig. 1. rl and r2 are the (amplitude) reflectivities
the solitary laser. Also, 7 p is the photon lifetime, 7inis the
of the laser facets, and r3 is the reflectivity of the external
diode cavity round-trip time, 7s is the carrier lifetime, J
mirror. We introduce the carrier density N ( t ) and the com-
is the pumping term, and R = nsP/7,,is the rate of spon-
plex electric field E(t) as the dynamic variables of the sys-
taneous emission into the lasing mode, with nsp being the
tem. N(t) is the average carrier density in the active region
spontaneous emission factor. The numerical values
and E(t) is the electric field at the inner right laser facet.
adopted for the various parameters, as appropriate for a
It is written as E(t) = Eo(t) exp [ j ( w o t + + ( t ) ) ] ,where
long wavelength InGaAsP laser diode [46], are given in
the amplitude Eo(t) and the phase +(t) are real. wo is the
Table I.
angular frequency of the solitary laser. For the case of
Equations (1)-(3) are written down for the case of a
weak to moderate feedback the dynamics of the laser sys-
Fabry-Perot laser, where the feedback parameter K is real
tem is governed by the following rate equations [2 11, [23],
and given by
[I11
d 1 K = (1 - r 3 r 3 / r 2 . (7)
- Eo(t) = - (G(N, E;) - l / ~ ~ ) E o (+t ) Eo(t - 7)
dt 2 7in K~ is the power reflected from the external cavity relative
R to the power reflected from the laser mirror. For a DFB
COS (U07 + A(t)) + ~
(l) laser diode K in general becomes complex and is given by
2 VCEdt)
a somewhat more complicated expression [ 111 which de-
d 1 K Eo(t - r ) pends on the details of the grating. Here, we shall assume
+(O = 2 aGdN - "1) --
Tin
~

Eo(t) K to be real and positive, but the results can easily be mod-
ified to include the phase of K. The condition for having
- sin (wor + A(t)) (2) weak to moderate feedback is K << 1 .
96 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO. 1, JANUARY 1992

TABLE I
LISTOF PARAMETER VALUES FOR LASERDIODE N

_
Parameter
_ ~
Symbol Value Unit t A A

Gain Coefficient 8.1 ’ 10-l~ m3s-’


Carrier density at transparency 1.1 . m-’
Camer lifetime 2.0 ns
Photon lifetime 2.0 PS
Laser cavity roundtrip time 8.0 PS
Linewidth enhancement factor 6.0 Fixed
Active region volume 1.0 . IO-’’ m’ point ’
r’
Confinement factor
Nonlinear gain coefficient
Spontaneous emission factor
2.5
0.3
.
2.2
m3
I Eo
Limit
cycle
Chaotic
attractor

Fig. 2. Schematic of attractors in the (Eo, A , N) space for a feedback level


in regime IV. Triangles ( A ) represent antimodes and crosses ( X ) represent
fixed points that have become unstable through Hopf bifurcations.
The Concepts of Attractors and External Cavity Modes
In this section we shall briefly introduce some of the
concepts from nonlinear dynamics and describe a typical wo)t (and hence A(t) = A, = (U,- w0)7), where N,, E,,
situation for a feedback level in regime IV. The back- and w, are constants. The dot indicates a stable fixed point
ground for the description will be discussed in detail in solution and is therefore an attractor. The triangles rep-
the following sections. resent unstable fixed point solutions of the saddle point
Before solving the rate equations (1)-(3) we need to type while crosses represent fixed point solutions that have
specify the initial conditions Eo(0),+(O), and N(O), and become unstable through Hopf bifurcations. None of these
due to the delay term also Eo(t) and +(t) in the time inter- unstable fixed points are attractors, but a cross is often
val -7 4 t I0 . The latter condition turns the system “close to” an attractor (cf. Fig. 2) which has been split
infinite dimensional since the time evolution actually de- off by a Hopf bifurcation at a lower feedback level. There
+
pends on the specified values for Eo and in a continuous may be crosses with no corresponding attractor, e.g., in
time interval, i.e., in an infinite number of points. The cases where the split-off solution has collided with the
phase space of the system is therefore the space of all basin boundary [l 11. As we shall see there can also be
pairs of suitably well-behaved functions Eo(t) and + ( t ) for fixed point solutions which have split off more than one
-7 It 5 0 combined with N ( 0 ) . attractor.
A solution to (1)-(3) describes a trajectory in, e.g., the It would be logical to define the attractors as the exter-
three-dimensional space (Eo, A , N). The trajectory ob- nal cavity modes of the system. However, for strong feed-
tained after transients have died away constitutes the so- back levels the attractors are generally not known, and it
called attractor. Apart from the feedback regime I there is therefore more practical to relate the external cavity
will in general be several coexisting attractors for a given modes to the fixed point solutions. In this paper we simply
feedback level. This is illustrated in the schematic ex- define the fixed point solutions which are unstable of the
ample in Fig. 2, which depicts a situation for a feedback saddle-point type (i.e., the triangles in Fig. 2) as the anti-
level in regime IV. The example illustrates some typical modes and the other fixed point solutions as the modes
attractors, such as the fixed point solution, the limit cycle, of the system. This definition ensures that the modes will
the torus, and the strange (or chaotic) attractor. actually be the attractors of the system for weak feedback.
The set of starting points in phase space from which the The antimodes can never be oscillating modes.
subsequent trajectories end up at a given attractor is called
the basin of attraction for the attractor. In the example of
Fig. 2 there are four attractors and the phase space is cor- Stationary Solutions and the Complex k-Plane
respondingly divided into four basins of attraction. If The externally controlled parameters for the external
Langevin noise terms are added to (1)-(3) to represent cavity laser are the bias current J, the feedback level K ,
spontaneous emission noise the solution may switch be- and the external cavity round-trip time 7. Stationary (or
tween the different basins of attraction and in this way fixed point) solutions to (1)-(3) with feedback are found
simulate mode hopping. The attractor close to which the by substitution of Eo(t) = E,, N ( t ) = N,, and +(t) = Aw,t
solution spends most of its time is called the dominant = (U, - w o ) t . When solving for the stationary solutions
attractor. For weak to moderate feedback levels (say K < one may, to a good approximation, neglect the small con-
0.1) the dominant attractor is usually the attractor with tributions from nonlinear gain suppression and sponta-
average value of A closest to zero, i.e., with average op- neous emission (i.e., we use E = 0 and R = 0). The
tical frequency closest to the frequency of the solitary laser stationary angular frequencies w, are the solutions w = w,
w . (511. to the equation
The dot, the crosses, and the triangles in Fig. 2 repre-
sent fixed point solutions, i.e., stationary solutions to (1)-
(3) of the form N(t) = N,, Eo(t) = E,, and + ( t ) = ( U , -
U, = w + sin (a7 + arctan a). (8)
Tin
M ~ R Ket a / . : CHAOS IN SEMICONDUCTOR LASERS 91

The solutions to (8) for fixed U, are easily found by graph-


ical construction (cf. Fig. 3) and define the external cavity
modes or antimodes of the system for a fixed set of pa-
rameters. The stationary carrier density and field ampli-
tude corresponding to a given value for U,7 are
2K
N, = Nso/ - -cos wsr (9)
GN 7in

The classification of fixed point solutions into modes w


and antimodes requires a stability analysis. Before dis- Fig. 3 . Graphical solution for the fixed point solutions (modes and anti-
cussing the stability properties, however, we introduce the modes) of the external cavity laser system for given U,, cf. (8).
plane of the complex wavenumber k for the laser wave-
guide, which turns out to be very convenient in charac-
terizing the mode properties.
We introduce coordinates x , y given by
x + jy = 21Ak (11)
where 1 is the laser diode cavity length and Ak is the
change in wavenumber from the solitary value. Using the
expression21Ak = qn{Aw + j 1 / 2 ( 1 + ja)GNAN} [47],
where A w = w - w, and AN = N - Nsol, (11) implies
y = rinGNAN (12)
x + a y = 7inAu. (13) -0.02 -0.01 0.00 0.01 0.02
X
The variable y is proportional to the threshold gain, and
Fig. 4. Complex k plane forJ/J,, = 1.3 and 7 = 1 ns (Example 1 ) . Modes
for a = 0 the variable x would simply be proportional to and antimodes ( 0 ) are shown for K = 2 . (inner circle), 8 .
the relative oscillation frequency A w = w - wo. For non- and 1.5 . lo-’ (outer circle), with the phase tuned to give a mode on the
zero a, however, the lines of constant frequency are in- thick solid line of minimum linewidth, i.e., Bo = 0.
clined relative to the x axis, cf. (13).
Insertion of (8) and (9) into (12) and (13) for w = w,
shows that the fixed point solutions fulfill the equation of the zeros of the corresponding system determinant D(s),
where s = j Q , Q being the baseband frequency. A (com-
( x , y) = -K (sin w s 7 , cos w,r) (14) plex) zero s = so implies a characteristic time dependence
i.e., they lie on a circle with radius K in the complex k 0: exp (sot) of a small perturbation from the initial sta-

plane. An example of the complex k plane is shown in tionary solution. Thus, the criterion for stability to small
Fig. 4. perturbations is that all zeros of D(s) lie in the left half
(Re {s} < 0) of the complex s plane.
Two different kinds of instability can be identified. If
Stability Analysis
Analysis of the small-signal stability properties of the %= 1+ K 1 cos (wr + arctan a ) < o
external cavity laser system were presented in [48], [4], dw 7in
[47]. These investigations, however, did not include the (15)
effect of nonlinear gain suppression, which turns out to
be important for the stability properties. Below, we give for w = us,the determinant D(s) has a zero on the real
a short summary of the stability properties including non- axis in the right half s plane [4], [47]. This means that the
linear gain suppression, which, to a good approximation, stationary solution is an unstable fixed point of saddle-
leads to a rather straightforward modification of the ap- point type, i.e., it is an antimode. Introducing x and y by
proximative results of [48]. The approximative results are (14), the stability boundary dw,/du = 0 imposed by (15)
further compared with exact numerical results, and in par- is given by
ticular it is demonstrated that the complex k plane allows
a simple representation of the results. (16)
The stability properties of a particular stationary solu-
tion are as usual investigated by performing a linear ex- and solutions to the left of (16) are unstable. The region
pansion around that solution and analyzing the positions where we have this saddle-point instability is shown
98 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO. 1, JANUARY 1992

2.0
hatched in Fig. 4 , i.e., the fixed point solutions in this
region are the antimodes. According to (15) the antimodes
1.5
are the graphical solutions in Fig. 3 where the sinusoidal
curve is decreasing with frequency. From the figure it is
clear that there is an antimode between each pair of ad- $e: 1.0

jacent modes on the frequency axis. The stability bound-


0.5
ary (16) is the line where the modes and antimodes are
created and annihilated. For increasing r the modes and
-.-
antimodes are created in pairs through saddle-node bifur- !

0 1 2 3 4 5
cations at the upper intersection point between the line
(16) and the circle (14). The modes move along the circle
in clockwise direction, and the antimodes move in anti-
0.010
clockwise direction. They meet at the lower intersection
point where they are annihilated through inverse saddle- 2 0.008
node bifurcations. For decreasing r the sequence of events
is reversed. 4 0.006

In addition to the condition dw,/dw > 0 for a mode to 5


be stable, we also have the stability condition [48], [ 1 11
fi O'Oo4
0.002

0.000

for all values of Q satisfying

Q2 - U; = -cot
TR
.);( Fig. 5. (a) Normalized angular frequencies f l / w R where a zero of D(s)
may pass the imaginary axis and (b) minimum feedback levels to reach the
corresponding stability boundaries (17) versus normalized extemal cavity
roundtrip time. The parameter region of instability is shaded. J/J,,, = 1.3.
Here, wR and rR are the relaxation oscillation frequency
and the corresponding damping time for the solitary laser,
and are given by The boundary lines resulting from (17) and (18) (shown
as solid in Fig. 4) are approximations to the curves where
D(s) has zeros on the imaginary axis [48]. The exact
curves are shown as dashed in Fig. 4 and are in fact close
where Esor is the field amplitude of the solitary laser. to the lines given by (17). Upon variation of r the lines
Equation (18) has discrete solutions for Q with spacing perform a complicated motion in the complex k plane.
approximately equal to 21r/r, cf. Fig. 5(a). Each of the This is illustrated in Fig. 5(b) which shows the distances
solutions for Q yields a stability boundary line given by from origo to the various lines, i.e., it shows the mini-
(17). The shaded area in Fig. 4 is the instability region mum feedback levels needed to reach the various stability
where (17) is not satisfied. Passing the first boundary (17) boundaries. The lowest lying curve for given uRr thus
for increasing feedback level a complex conjugate pair of gives the minimum feedback level to reach the first sta-
zeros of D(s) moves into the right half s plane at s = h'Q.bility boundary. This boundary in general corresponds to
This means that the fixed point solution (E,, As,N,) be- the solution Q of (18) closest to oR,cf. Fig. 5(a). It fol-
comes unstable and a stable periodic (limit cycle) solution lows from (18) that Q = oRfor
is created where the field amplitude, phase delay, and car-
U R T = ?r + 2rp, p integer (20)
rier density oscillate with angular frequency Q around the
stationary values E,, As = (U, - U,) r , and N,. The insta- and in this case (17) shows that the distance to the bound-
bility region where (17) is not satisfied has correspond- ary line becomes [49], [50]
ingly been called the region of feedback-induced intensity
pulsations [48] or the region of dynamic instability [4].
K = Tin/(27RJ1 (r2) + (21)
The birth of a stable limit cycle solution from a stable which is shown as the dashed line in Fig. 5(b). For long
fixed point solution is known as a Hopf bifurcation [35]. extemal cavities, w R r >> 1 , (21) is seen to provide a
Notice, that after such a bifurcation the solution remains good approximation to the value for K at which instability
in the neighborhood of the old fixed point solution which sets in, independently of the precise value for w R r . Re-
is now unstable, and we may therefore still associate the cently, (21) was suggested as a simple analytical expres-
solution with the external cavity mode. This is in contrast sion for the onset of coherence collapse by Helms and
to the other stability boundary (16) where a stable and Petermann [50]. However, we should like to point out
unstable solution will collide and disappear, and the so- that the feedback level given by (21) only marks the start-
lution will be swept away to some distant attractor. ing point of a sequence of instabilities which eventually
M@RK et al.: CHAOS IN SEMICONDUCTOR LASERS 99

leads to a chaotic state at a somewhat higher feedback


level, as shown below. The distinction between onset of
relaxation oscillations and onset of coherence collapse was
also made in [4] and in a recent experimental paper by
Woodward et al. [5]. Notice in Fig. 5(b) that for pwR7 <
?r the stability region pertains to very high feedback lev-
els, thus explaining the observed stability of lasers with
very short external cavities [51]. N
The number of modes is determined by the parameter
C = ~ ( 7 / 7 ~J "1)+;;"[52]. Thus, for C < 1 only one
mode exists, while for C > 1 there exists a region of
feedback phase where two or more modes exist. For C
>> 1 the total number of modes is approximately given
by C/?r + 1. The condition C = 1 has a simple interpre-
tation in the complex k plane (cf. Fig. 4). It corresponds 0
+
to the case where the circle x 2 y 2 = K~ just touches the A
instability line given by (16). The inner part of the circle Fig. 6. Example of potential V A ) (upper figure) and modes ( 0 ) and
antimodes (A)in the (A, N ) plane (lower figure).
represents the feedback regime I of single mode opera-
tion.
In the presence of several possible attractors, the issue
respond to the bottoms of the valleys of the potential while
of global stability becomes important. Thus, spontaneous
the antimodes correspond to the hill tops. The noise term
emission quantum noise may induce transitions between
in (22) will induce a switching of A between the valleys.
the attractors, and the question is which solution is dom-
The switching corresponds to mode hopping and from the
inant. It is usually assumed that a laser diode oscillates in
potential model one can derive transition times which are
the mode with lowest carrier density since this corre-
in surprisingly good agreement with experiments [27].
sponds to the lowest gain necessary for lasing to occur.
The dominant mode will be the mode with the deepest
According to (12) this would be the lowest lying mode on
valley and it can easily be seen from (23) that this will
the circle in Fig. 4. However, it has been demonstrated
always be the valley closest to A = 0. Thus, the dominant
experimentally [3], [53], [27], and numerically [23], [24]
mode is the mode with frequency closest to the frequency
that a laser diode with weak to moderate feedback pre-
of the solitary laser. The linewidth A V is in [47] shown
dominantly oscillates in the external cavity mode with
to be given by A v / A v , = (dpw,/dw)-2 which, using (15)
lowest linewidth. In [24], [27], [28] this quite unexpected
and (23), can also be expressed as A v / A v , =
behavior was explained from a potential model for phase
( 7 d 2 V / d A 2 ) - 2 .This shows that the linewidth of a specific
fluctuations.
For a feedback level K < 7in/(27~ w), where the
external cavity modes are all stable fixed points, the dy-
mode is inversely proportional to the square of the poten-
tial curvature at that point [27]. Since the mode with deep-
est valley also has the largest potential curvature, the
namics of mode hopping is governed by the equation [27]
dominant mode is also the mode with minimum line-
d width. The expression
- A(t) = -
dt dA
+ FA(f)
7in
where
1 shows that the level curves of constant linewidth are lines
V(A) = - - K cos (e, + A ) . (23) parallel to (16). The linewidth is infinite along (16) and
27 Tin
decreases rapidly with increasing distance to (16). The
FA(t) is a Langevin noise function which satisfies the cor- line of minimum linewidth is therefore perpendicular to
relation relation ( F A ( t ) F A ( t ' ) ) = 4nAvO6(t- t ' ) , where (16) and given by (13) for Apw = 0, i.e., it is identical to
A V , is the linewidth of the solitary laser. 0, is the phase the line where the modes have the frequency of the soli-
tuning tary laser. According to (8) the minimum linewidth (or
Apw = 0) is obtained for the phase tuning eo = 0.
8, = w,7 + arctan a. (24) The potential model breaks down for higher feedback
Equation (22) is analogous to the equation of motion for levels ( K > J1+;;"))where the modes become
a particle moving with strong friction in a potential V(A) unstable. However, numerical calculations show that even
and exposed to a fluctuating force FA@).An example of for feedback levels up to K = 5 * lo-* the dominant at-
the potential (23) is shown in Fig. 6 for a phase tuning 8, tractors will be those with average values of A closest to
= - 1.O. The figure also shows the fixed point solutions zero. It is not clear to us what happens for even stronger
in the (N, A ) plane. In this plane the circle (14) is trans- feedback where (1)-(3) are no longer valid. In the follow-
formed into an ellipse [34]. We notice that the modes cor- ing section we will investigate the evolution of the mini-
100 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO. I, IANUARY 1992

mum linewidth mode, i.e., the mode on the thick solid


line in Fig. 4 with phase tuning 8, = 0. Whereas the tran-
I
sition from a fixed point solution to a limit cycle is pre- 0
dicted by the stability analysis, this analysis tells nothing b. 0-
about the future evolution of the limit cycle for increasing 3
. I 4

feedback level. v)
G -5-
%
111. NUMERICAL
RESULTS k
The rate equations (1)-(3) are solved numerically by -10-
using a technique based on Gear’s method for stiff sys-
tems. As discussed above, a solution to (1)-(3) describes
3
a trajectory in (Eo, A, N) space which approaches an at- 0.000 0.005 0.010 0.015 0.020 0.025
tractor as the transients die away. It is convenient to char-
Feedback level K
acterize the attractor by recording its intersection points
(a)
with a transverse plane. The resulting two-dimensional
plot is denoted the Poincark section [35]. For the results l5 i
to be presented, we have chosen the plane E = Esol as the 10 -I ’I
intersection (Poincark) plane, where Esoris the field am-
plitude of the solitary laser. Thus, the Poincar6 section
shows the transverse extent of the attractor in the (N, A)
plane for E = Esol. Two examples of bifurcation se-
quences will be discussed.

Example 1
The first example is for J / J t h = 1.3 and T = 1 ns. The
complex k plane for this case is shown in Fig. 4. Fig. 7 -15 1 I 1 I I 1
shows bifurcation diagrams obtained for increasing feed- 0.000 0.005 0.010 0.015 0.020 0.025
back level along the line of minimum linewidth. The dia- Feedback level K
grams show the values of normalized carrier density (b)
N/N,,l - 1 [Fig. 7(a)] and phase delay A [Fig. 7(b)] in Fig. 7 . Bifurcation diagrams for the minimum linewidth mode showing (a)
normalized camer density N/N,,,- 1 and (b) phase delay A. (Example
the Poincark section. Notice that only intersection points 1).
corresponding to traversal through the Poincark plane in
the direction of decreasing Eo are recorded, i.e., “mir-
ror” points are rejected. Fig. 8 shows examples of pro- quencies [Figs. 8(b), 9(b), and 1l(b)j. The new frequency
jections of the three-dimensional attractors onto the (Eo, that comes into play is related to but slightly lower than
A) plane for different values of K, and Figs. 9 and 10 show the inverse 1/ T of the external cavity round-trip time. The
the corresponding Poincark sections. Finally, in Fig. 11 quasi-periodic nature of the solution is clearly evidenced
we show examples of the actual time evolution, since pic- by the closed Pioncark section [Fig. 9(b)], which shows
tures of the attractors in phase space provide no infor- that the trajectory lives on a two-dimensional closed sur-
mation on the time scales of the dynamics. face (a torus). The bifurcation from a limit cycle solution
No intersections with the Poincar6 plane exists when to a quasi-periodic solution is sometimes denoted a sec-
the solution is a stable fixed point, but for K = 2.4 lop3 ondary Hopf bifurcation, because in the Poincark plane it
relaxation oscillations become undamped and give rise to corresponds to a fixed point being replaced by a closed
a self-sustained periodic (limit cycle) solution with fre- curve. The bifurcation can be analyzed by performing a
quency Cl/27r = 2.38 GHz [Figs. 8(a), 9(a), and ll(a)]. stability analysis around the periodic limit cycle solution
The limit cycle solution leads to a single intersection point using the technique of Floquet multipliers [35]. However,
with the Poincare plane. This agrees with the stability in our case such an analysis has no predictive power, since
analysis since for K = 2.4 a pair of complex con- it requires the availability of the limit cycle solution,
jugate zeros of the system determinant D(s) moves into which has to be computed numerically anyway.
the right half of the complex s plane at s/27r = 2.38 For K = 1.21 * the limit cycle of order six bifur-
GHz. Notice, that the relaxation oscillation actually cor- cates to a torus of order six [Figs. 8(e) and 9(e)] as evi-
reponds to a “slow” modulation of the true optical car- denced by the closed curves obtained in the Poincark sec-
rier, and it leaves the mean optical frequency wo + tion [Fig. 9(e)]. The two rightmost nonclosed curve
( & / d t ) almost unaffected. segments in Fig. 9(e) arise because the torus bends over
For K = 7.4 * lop3the limit cycle solution bifurcates very close to the Poincark plane, cf. Fig. 8(e).
to a quasi-periodic solution with two incommensurate fre- For K = 1.21 * lop2the limit cycle of order six bifur-
M@RK er al.: CHAOS IN SEMICONDUCTOR LASERS 101

40 I I 60 I I
3.0 - 4.0 -

2.0 -

-20-
-2.0 -

1 -
0.00
-4.5 -4.0 -3.5 -3.0
50 -
Carrier density
;
g
20-
10-
40-

20-
(a)

g 00- 00-
g -10-
-20-
a
-20 - & 20-
4

4.0 60
2% 0°-

3.0

;
3
2o
1.0
40

20
-20 I
-7.5
I
-5.0
I
-2.5 0
g 0.0 Carrier density (W3)
2 -1.0
00

-20
(b)
-20 Fig. 10. Poincar6 sections. (a) Blowup of the framed part of Fig. 9(f)
I ( K = 1.2279 . lo-’). (b) K = 1.25 . (Example 1).
-30
0- 3 -‘’’ -0.4 -02 0.0 02 0.4
Electric field Electric field

Fig. 8. Projections of attractors. (a) K = 4 . (b) 8 . (c)


8.72 . lo-’, (d) 1.16 . (e) 1.2138 . ( f ) 1.2279 . (Ex-
ample 1).

-ol ’ 0 4
3.0-

20 -

10- .
0.0-

-10-

-50 -45 -40 -35 -30 -70 - 6 0 -50 -40 -30 -20 -10 00

(a) (d) o ~ ~ o
Time [nspt] rime I-[
-ol 40
(b) (4
Fig. 1 1 . Time evolution of electric field amplitude. (a) K = 4 . (b)
8. (c) 1.16 . lo-’, (d) 1.5 . (Example 1).
10

00
cates to a torus of order six [Figs. 8(e) and 9(e), as evi-
-10
- denced by the closed curves obtained in the Poincark sec-
-2 0
-70 -60 -50 -40 -30 -20 -10 00 tion [Fig. 9(e)]. The two rightmost nonclosed curve
segments in Fig. 9(e) arise because the torus bends over
(e)
very close to the Poincark plane, cf. Fig. 8(e).
10 , I - O 4
For K = 1.22 lo-* the torus bifurcates to a chaotic
attractor [Figs. 8(f) and 9(f)]. This is evidenced by the
multisheeted appearance of the Poincark section (see
blowup in Fig. 10(a)], which signalizes the presence of

- -50
1 4 5
5 -40
1
-35 -30
1 -20
-70
1 ,
-60
-,
-50
, ,
- 4 0 -30 -20
,
-10
, I
00
the basic stretching and folding processes underlying cha-
otic attractors [35].
Just after the onset of chaos the attractor stays in the
Carrier density Carrier density
neighborhood of the old torus solution, but for slightly
(c) (f) larger K [Fig. 10(b)] the points in the Poincark section
Fig. 9 . Poincar6 sections of the attractors shown in Fig. 8. (Example 1). start to “diffuse” inbetween the previously [Fig. 9(f)]
102 IEEE JOURNAL OF QUANTUM ELECTRONICS. VOL. 28, NO. 1, JANUARY 1992

isolated ‘‘bands. ” Thus, the attractor suddenly starts to 0 010


take up a larger volume in phase space, and the “band”
structure quickly gets lost. The time trace well inside the
0 005
chaotic region [Fig. ll(d)] of course shows completely
erratic dynamics.
The bifurcation diagrams [Fig. 71 show that even after h 0000
onset of chaos, there may be parameter windows of pe-
riodicity (high order frequency locking). Also, the bifur-
-0 005
cation diagram [Fig. 7(b)] for the phase delay A shows
that even when the chaotic attractor extends over several
external cavity modespacings, there is still a reminiscence -0 010
of the old mode structure. -0010 -0005 OOOO 0005 0010
The bifurcation sequence just described for increasing X
feedback level shows that the system undergoes a quasi- Fig. 12. Complex k plane. Modes and antimodes ( 0 )are shown for K =
7 . Solid curves in the shaded area are exact stability boundaries.
periodic route to chaos interrupted by frequency locking. (Example 2).
In other examples, i.e., for other values of bias current
and external cavity length, we have observed essentially
the same pattern. However, the system may also go di-
rectly from a torus to chaos without frequency locking, m
-

and the torus itself may also undergo a period doubling. 0


b. 5- 0
We have seen no frequency locking of fractional order
which is probably due to the fact that higher harmonics 2
of the relaxation oscillation are at high frequencies and $
a 0-
-5
are strongly damped. In Example 2 below, we will dis- L
cuss a different transition sequence and additional com- .9
L -5- -10
plications that arise due to the presence of multiple at- L
tractors. 8
-10 1 I I 1 -15
0.005 0,010 0.015 0.020 0.025
Example 2 Feedback level K
Fig. 13. Bifurcation diagrams for the minimum linewidth mode. Upper
The second example is for J / J t h = 2 . 0 and r = 2 ns. part (right axis) is obtained for increasing K and lower part is obtained for
Fig. 12 shows the complex k plane and Fig. 13 shows decreasing K . The traces are separated vertically for clarity. (Example 2).
bifurcation diagrams for the minimum linewidth mode.
The upper trace in Fig. 13 (right axis) was obtained for
increasing feedback level and shows the creation of a limit
cycle (at K = 6.0 - which evolves into a torus (at K
= 8.6 - and then into chaos (the “explosion” at K
-
= 1.15 lo-’). The transition to chaos at play here will
be discussed later (see Figs. 15 and 16), but for the mo-
ment we will concentrate on the discontinuous jump seen
for K 2: 1.20 which signalizes that the solution
jumps to another attractor. The second attractor under-
goes a usual quasiperiodic route to chaos, involving fre-
-1.2 -0.8 -0.4 0.0 0.4 0.8 1.2
quency lockings of order eight and sixteen. Tracing back
the second attractor, cf. lower trace in Fig. 13 (left axis), Phase delay A/(27~)
shows that it coexists with the first attractor and is created Fig. 14. Projection of coexisting attractors for K = 7.0 . lo-’. The dot
indicates a stable mode and the crosses indicate unstable modes. (Ex-
as a limit cycle solution at K = 6.4 - The “old” ample 2).
chaotic attractor may also exist beyond the jump at K =
1.20 l o p 2 , but due to the self-generated noise it is nu-
merically hard to avoid switching to a nearby solution with stable solution, i.e., constitutes an attractor. The other
more “damped” trajectories. two modes (the crosses) are unstable but still have attrac-
Fig. 14 shows projections of the various attractors cor- tors associated with them, which are limit cycle solutions
responding to the case of Fig. 12 ( K = 7.0 * l o p 3 , 8, = oscillating around the positions of the unstable modes. It
0). The attractors are projected onto the plane of normal- is important to notice that there are two limit cycle solu-
ized carrier density N / N S o l - 1 and phase delay A. Three tions associated with the middle (dominant) mode. These
modes exist for these parameters and are shown as dots are the two coexisting attractors for which bifurcation dia-
and crosses in Fig. 14. However, in agreement with Fig. grams were shown in Fig. 13. Whereas the mean values
12 only the leftmost of these (the dot) corresponds to a of phase delay (and, hence, optical frequency) and carrier
M@RK er al.: CHAOS IN SEMICONDUCTOR LASERS I03

density are nearly equal for the two attractors, their relax- 2 0 ,

ation frequencies clearly differ. Thus, the attractor marked


1 has a period (measured as the time between consecutive
intersections of the Pioncar6 plane) which corresponds to
a frequency of 4.24 GHz, while attractor 2 has a fre-
quency of 3.83 GHz. The rightmost limit cycle has a pe-
riod equal to that of attractor l.
A small-signal stability analysis shows that the “first”
limit cycle solution (marked 1 in Fig. 14) is created by a
normal Hopf bifurcation at K = 6.0 * where a pair 20, I , I

15-
of complex conjugate zeros of the system determinant D(s)
moves into the right half of the complex s plane at s/27r ;
8 05-
lo-

= fj 4.24 GHz. The old stable fixed point bifurcates g 00-

to an unstable fixed point and a stable limit cycle with 2 05-

oscillation frequency 4.24 GHz. For a slightly larger -10-


- 51 -
feedback level, K = 6.3 * (see Fig. 12), another pair -80 -60 -40 -20 00 -80 -60 -40 -2.0 00

of complex conjugate zeros moves into the right half s Carrier density Carrier density
plane at s / 2 n = fj 3.83 GHz. This indicates a second (b) (4
Hopf bifucation from the now unstable fixed point which, Fig. 15. Poincare sections near the “explosion” in Fig. 13. (a) K =
in general, leads to the creation of an unstable limit cycle. 1.13 . lo-’, (b) 1.1675 . lo-’, (c) 1.180 . lo-*, (d) 1.181 . lo-’.
For K > 6.4 - however, direct numerical integra- (Example 2).
tion shows the presence of a stable limit cycle with a fre-
quency of 3.83 GHz that is centered around the unstable
fixed point solution, cf. Fig. 14 (K = 7.0 * lop3). This all the points in the Poincar6 section [Fig. 15(b)] which
strongly suggests that the stable limit cycle bifurcates off lie within the slice -ANc,,/2 < N - N,,, < AN,,,/2.
the unstable limit cycle for K = 6.4 - By analyzing In Fig. 16(a) we have plotted the values of phase delay A
a model system of two coupled oscillators (p. 398 in [54]) for points in this slice versus the values of A at the pre-
we have found that a second Hopf bifurcation in general vious intersection with the Poincar6 plane E = Esol.Since
leads to the birth of an unstable limit cycle which may, the time T between consecutive intersections of the Poin-
however, quickly bifurcate into a stable limit cycle and car6 plane E = Esolis approximately constant, this means
an unstable torus. This might be the way the second at- that Fig. 16(a) essentially shows A(?) versus A(t - T),
tractor in Figs. 13 and 14 is created. where the points A(t) are taken at times t fulfilling E(t) =
The birth of another attractor with different relaxation Esol and -AN,,,/2 < N(t) - N,,, < AN,,,/2. We
oscillation frequency is clearly observed in cases like this have used Ncu,/Nso, - 1 = -5.65 and AN,,,-
where two sets of zeros of D(s) move into the right half s = 0.0125 The closed curve obtained in Fig. 16(a)
plane for almost the same K, cf. Fig. 12. Referring to Fig. demonstrates that the attractor corresponding to Fig. 15(b)
5(b), this is expected to happen for values of J / J , , and 7 is a three-dimensional torus, i.e., it corresponds to a
chosen so that wR7 2: p 27r, p integer. For large values quasi-periodic solution with three incommensurate fre-
of p (long cavities, e.g.) this is not a severe requirement, quencies (a two-frequency torus would lead to two points
though, cf. Fig. 5(b). Even if the birth of a stable solution instead). The width of the curves in the secondary Poin-
from an unstable mode is quite unexpected, it will be car6 section decreases for decreasing value of the slice
shown that this solution is also clearly identified in the width AN,,,. Notice, that our introduction of a new vari-
experiments, where the quantum noise induces random able A(t - T) is allowed by the infinite dimensional na-
transitions between the two attractors (see Fig. 20). ture of the feedback system. Until quite recently, three-
The dynamics involved in the “explosion” of the first frequency quasi-periodic solutions were believed to be
-
attractor for K = 1.15 lop2is analyzed in Figs. 15 and nongeneric [55], and have not previously, to our knowl-
16. The closed curve obtained in the Poincar6 section for edge, been found in other laser models.
K = 1.13 lo-* [Fig. 15(a)] shows that just before the Fig. 15(c) and (d) show the Poincark sections for
“explosion” the attractor is a torus. However, the Poin- slightly larger values of K. Fig. 15(c) indicates some de-
car6 section quickly gains a certain width, and for K = gree of frequency locking on the three-torus (frequency
1.1675 - [Fig. 15(b)] the Poincar6 section looks locking between two frequencies on a three-torus of course
more like a projection (with shadows, e.g.) of the attrac- leads to a normal two-torus), but the curves have a blurred
tor than the actual transverse cut it represents. This sug- appearance, which we believe is an indication of chaotic
gests that we are dealing with an attractor of more than behavior. This is even more pronounced in Fig. 15(d), for
two dimensions (the dimension of a torus arising due to which we have shown the secondary Poincar6 section in
two incommensurate frequencies). We have investigated Fig. 16(b). Fig. 16(b) clearly indicates a multisheeted
this by performing a secondary Poincark section. Thus, structure, which is a good indication that we are dealing
for a specific carrier density, N = N,,,, we have recorded with a chaotic attractor.
104 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO. 1, JANUARY 1992

PZT Atten. BS FP Det.

&or DFB LD lsolators II I

Delay

20
Fig. 17. Experimental setup. PZT = Piezo-electric transducer; Atten. =
neutral density filter; BS = beamsplitter; FP = Fabry-Perot interfer-
ometer; Det. = photodetector; Osc. = oscilloscope; Amp. = amplifier.
a
00

--<'
output current from the photodiode (10 GHz bandwidth)
-1 0
is mixed with at strong local oscillator signal with fre-
-12 -08 -04 00 04 08 quency fLo in a nonlinear mixer. The output from the
mixer contains the frequency difference signal, and this
signal is then studied directly in the time domain with a
Fig. 16. Secondary Poincar6 sections taken at N / N , , , - 1 = -5.65 . real-time digital storage oscilloscope (500 MHz band-
lo-' of the attractors in Fig. 15(b) and (d). (a) K = 1.1675 . IO-' and width). The noise level of our detection system (including
(b) 1.181 . lo-'.
amplifiers) is much lower than the inherent level of quan-
tum noise in the laser. Phase portraits of the dynamic be-
IV. EXPERIMENTAL RESULTS havior were obtained by splitting the signal after the mixer
Fig. 17 shows the experimental setup. A 1.3 pm dis- in two arms and delaying one of these by a quarter of the
tributed-feedback laser diode from Fujhitsu was used for oscillation period. The resulting single-shot x-y traces for
the experiments. The laser was temperature stablized to various feedback levels are shown in Fig. 18. The fre-
within 0.01"C. The distance to the (flat) external mirror quency of the local oscillator was fixed at fLo = 4.088
was accurately controlled with a piezo-electric trans- GHz and change of this (or, equivalently, the electric de-
ducer, thus allowing control of the phase of the feedback lay time) only led to attractors of more elliptical shape.
light. A variable attenuator inserted in the external cavity The first trace in Fig. 18 is for a low feedback level, K
was used to vary the feedback level, and the absolute level = -45 dB, where the laser still operates stably in the
of feedback was calibrated by identifying the level of at- minimum linewidth mode. Even though the linewidth is
tenuation at which several modes start to exist (corre- narrowed considerably in comparison with the solitary
sponding to the previously discussed condition C = 1). laser, the intensity noise level is somewhat larger than for
The optical spectrum of the laser was monitored with a the solitary laser. The origin of the laser diode noise is
confocal Fabry-Perot interferometer (free spectral range mainly spontaneous emission quantum noise. While the
(FSR) of 2 GHz), and the phase of the feedback light was laser linewidth reflects low-frequency noise, which is
controlled to assure oscillation in the minimum linewidth damped for weak feedback from an external cavity, the
mode (i.e., at the solitary laser frequency). intensity noise level, as measured here, is larger with
Experiments were performed with external cavity feedback due to less damping of the relaxation oscilla-
lengths of 16 and 30 cm. In order to avoid the additional tions.
complications arising due to multiple attractors with dif- At K = -43 dB in Fig. 18 a noisy limit cycle has
ferent relaxation oscillation frequencies, the basic route evolved which then grows in size with increasing feed-
to chaos is better investigated with a relatively short cav- back level ( K = -39 dB). The basic frequency of the self-
ity. On the other hand, a long external cavity is appropri- sustained relaxation oscillations is 4.20 GHz, which
ate to investigate the predicted competition between re- means that the difference signal has a frequency of = 110
laxation oscillations. MHz. For K = -34 dB the limit cycle is distorted to a
degree that cannot just be explained as being due to ran-
Route to Chaos dom quantum noise, and for K = -31 dB the trajectory
For these experiments the external cavity length was 16 has a completely different character which is typical of
cm, corresponding to a round-trip time 7 = 1.07 ns. chaotic behavior.
Due to the very short time scale (=0.25 ns) on which The heterodyne technique used in obtaining these re-
the basic relaxation oscillations occur, it is difficult to sults of course in some sense obscures the picture when
measure the time dependent behavior directly. Instead, we more than one frequency is present in the basic signal.
have employed an RF-heterodyne technique where the However, the results of Fig, 18 are consistent with the
M ~ R Ket ai.: CHAOS IN SEMICONDUCTOR LASERS 105

-45 dB -43 dB -39 dB

<'
-34 dB -31 dB
0 1 2 3 4 5
Fkequency (GHz)
(a)

Fig. 18. Experimental phase portraits for various feedback levels. J/J,h
1.83, r = 1 ns.
=
.d
4
E -r
5.2

5.0
I I I1

predicted quasiperiodic route to chaos. We should also 4,


like to remark, in general, that in going from quasi-peri- -44 -40 -36 -32
odic behavior to chaos, all the associated peaks in the Feedback level IC (dB)
spectrum of the signal are expected to broaden. In that (b)
sense it may be appropriate to study the (downconverted) Fig. 19. (a) Measured intensity spectrum and (b) frequency ratiof,/f, ver-
dynamics within a certain bandwidth of one of the fre- sus feedback level with identification of regimes I (limit cycle), I1 (quasi-
periodic), I11 (frequency locked), and IV (chaotic). J / J , h = 1.84, T = 1.07
quency peaks. ns.
In Fig. 19 we report a much more accurate test of the
quasi-periodic route to chaos by examining the evolution
of the intensity noise spectrum as a function of feedback In other examples we have observed essentially the
level. For these measurements the signal from the pho- same route to chaos as reported in Fig. 19, but without
todiode was amplified and analyzed with a microwave an intermediate region of frequency locking. Such a route
spectrum analyzer. The bandwidth of the detection sys- is of course also possible and has been observed in our
tem including amplifiers was ~ 4 . GHz. 5 numerical simulations. In the absence of any noise, a route
Fig. 19(a) shows an example of the observed spectrum to chaos interrupted by frequency locking seems more
of the laser light intensity. The two basic peaks C and R likely, but we note that the relatively high level of spon-
are situated at frequencies& andf,, where& = 7 - l andf, taneous emission noise may act to narrow the feedback
is close to the relaxation frequency of the solitary laser. region of frequency locking. Also, frequency locked states
The specific example of Fig. 19(a) demonstrates a fre- at higher locking ratios are more easily perturbed and as
quency locked state where all peak frequencies are mul- such more difficult to observe experimentally.
tiples of the low-frequency peak at fc. Specifically, f,/& By adjusting the numerical parameters to fit the bias
= 5.0, implying that we see the spectrum of a frequency level, external cavity length, and relaxation frequency of
locked state of order 5 . Halfway between the strong peaks Figs. 18 and 19, our numerical simulation shows a route
there are small bumps which might be attributed to a pe- to chaos which is quite close to the experimental obser-
riod doubling of the frequency locked solution, cf. Figs. vations (frequency locking of order 5 , and bifurcations
7 and 13. taking place at approximately the same feedback levels).
In Fig. 19(b) we plot the frequency ratiof,/f, versus K However, we would not claim a quantitative agreement
and four different regimes are identified. Regime I, which between theory and experiment since the various param-
starts at K = -44 dB, is the limit cycle regime charac- eters involved in laser diode modelling are not really
terized by a strong relaxation peak at R and a weak and known with an accuracy that permits so.
broad peak at C. The latter peak arises due to noise ex-
citation of the resonance that becomes undamped as a Hopping between Relaxation Oscillations
quasi-periodic solution is created from the limit cycle so- For these experiments the external cavity length was 30
lution. Thus, moving into regime I1 of quasi-periodicity cm, corresponding to a roundtrip time 7 = 2 ns.
the amplitude of C increases strongly. Regime I11 is char- Fig. 20(a) shows the time averaged optical spectrum as
acterized by a plateau in f r / f c at the integer value 5 . It is observed with a planar Fabry-Perot with 12 GHz FSR. In
therefore a regime of frequency locking. In the chaotic addition to the central peak A which reflects the oscillat-
regime IV the peaks quickly broaden and are hardly re- ing (minimum linewidth) external cavity mode, there are
solvable at high feedback levels. These observations thus two relaxation oscillation sidebands B and C (and their
provide strong evidence for a quasi-periodic route to chaos symmetrical counterparts). The peaks B and C are dis-
interrupted by frequency locking. placed from the central peak by 4.36 and 4.79 GHz, cor-
106 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO. 1, JANUARY 1992

responding to the positions of two strong peaks in the in-


tensity noise spectrum. The widths of all the peaks in the
optical spectrum Fig. 20(a) are limited by the 250 MHz
resolution of the Fabry-Perot interferometer. The peaks
B and C are identified as relaxation oscillation sidebands
of A by their behavior for increasing bias current. As
mentioned above, the relaxation resonances that exhibit
the least damping, correspond to the solutions to (18)
which have frequencies closest to wR. As we increase the
bias current, wR increases as ( J - J,h)’/*. We will there-
fore for increasing bias current see relaxation oscillation -6 -3 0 3 6
peaks rising and reaching a maximum when (20) is ful-
filled and then dying away when the next solution to (18)
Optical Requency (GHz)
is taking over [56]. (a)
In order to time resolve the dynamics underlying the
time-averaged spectrum Fig. 20(a), we used two Fabry-
Perot interferometers (12 GHz FSR) as frequency dis-
criminators, i.e., as filters with transmission around a
specific center frequency only. The two upper traces in
Fig. 20(b) result when the Fabry-Perot interferometers
are set at the frequencies of the two sidebands B and C. Li
The anticorrelated nature of the traces clearly show that 4
Q)
the laser jumps randomly between the two bands, thus
corresponding to random jumps between two periodic (or Hc 2 A
possibly quasi-periodic) solutions with different oscilla-
tion frequencies. The two lower traces result when one of
the Fabry-Perot interferometers is set at the central peak
A and the other is set at the first sideband B. It is seen that
the amplitude of the central peak also changes in response
to the jumps between sidebands. Compared to the case of
modejumps [27] (which may also be studied using this
Fig. 20. (a) Time-averaged optical spectrum. The frequency of the soli-
setup and result in a nearly 100% modulation of the sig- tary laser corresponds to 0 GHz. (b) Time dependence of peaks B and C
nal) these changes are however very small (notice that a (upper two traces) and A and B. Dotted lines indicate zero intensity level
common zero for the two lower traces is used) and are (common for two lower traces). JIJ,,, = 2.01, T = 1.97 ns, K~ = -37 dB.
mainly attributed to a difference in spectral linewidth be-
tween the two attractors or a small change in the average
optical frequency. Furthermore, jumps between different levels (in the range of -50 to -30 dB) and external cav-
external cavity modes would lead to two central peaks in ity lengths on the order of a few tens of centimeters. Even
the time averaged spectrum. This proves that the attrac- with these restrictions, the system allows extremely rich
tors belong to or arise from the same external cavity mode. and complicated dynamics. Two main issues have been
When the two attractors have just been created, at a adressed.
lower feedback level, the average time between attractor First, it was demonstrated that for increasing feedback
jumps is very small, then increases for an increasing feed- level the laser undergoes a quasi-periodic route to chaos
back level, attains a maximum, and finally drops rapidly. which may be interrupted by frequency locking. This, we
The observation of an initial increase of average jumping believe, explains the previously observed transition [3]
time we attribute to an increased spatial separation be- from a feedback regime of single mode narrow linewidth
tween the two attractors. The subsequent decrease in operation to a regime of coherence collapse [2]. Numer-
jumping time probably reflects that the attractors become ically constructed Poincark sections demonstrated that
chaotic, so that the trajectories more likely are taken close within the regime of coherence collapse, the laser dynam-
to the basins of attraction of the attractors. ics display the typical characteristics of chaos. The exis-
tence of a three-frequency quasi-periodic attractor was
V. CONCLUSION predicted theoretically, but has not been observed exper-
imentally. In general we should like to remark, that the
We have presented a detailed theoretical and experi- relatively high level of spontaneous emission quantum
mental investigation of the nonlinear dynamics of a semi- noise in laser diodes may prevent the observation or even
conductor laser diode with delayed optical feedback. Our existence of such extremely complex and delicate attrac-
study has been restricted to the case of moderate feedback tors.
M$RK et al.: CHAOS IN SEMICONDUCTOR LASERS 107

Second, the coexistence of two attractors associated backscatter-induced linewidth reduction,” Electron. Lett., vol. 21,
pp. 1008-1009, Oct. 1985.
with the same external cavity mode but having different K. Ikeda, H. Daido, and 0. Akimoto, “Optical turbulence: Chaotic
relaxation oscillation frequencies was demonstrated. In an behavior of transmitted light from a ring cavity,” Phys. Rev. Lett.,
experimental situation, spontaneous emission noise leads vol. 45, pp. 709-712, Sept. 1980.
to random jumping between the two attractors, thus re- M. Le Berre, E. Ressayre, A. Tallet, and H. M. Gibbs, “High-di-
mension chaotic attractors of a nonlinear ring cavity,” Phys. Rev.
sulting in two strong peaks in the intensity noise spec- Lett., vol. 56, pp. 274-277, Jan. 1986.
trum. The origin of the second attractor was identified as J. D. Farmer, “Chaotic attractors of an infinite-dimensional dynam-
ical system,” Physica, vol. 4D, pp. 366-393, 1982.
a second Hopf bifurcation from an unstable external cav- J. M$rk, P. L. Christiansen, and B. Tromborg, “Limits of stable
ity mode. However, the details of the bifurcation need operation of AR-coated semiconductor lasers with strong optical feed-
further clarification. back,” Electron. Lett., vol. 24, pp. 1065-1066, Aug. 1988.
R. Lang and K. Kobayashi, “External optical feedback effects on
semiconductor injection laser properties,” IEEE J. Quantum Elec-
tron., vol. QE-16, pp. 347-355, March 1980.
ACKNOWLEDGMENT K. Petermann, Luser Diode Modulation and Noise. Dordrecht: Klu-
wer Academic, 1988.
We wish to thank M. P. Sqirensen, Laboratory of Ap- N. Schunk and K. Petermann, “Numerical analysis of the feedback
plied Mathematical Physics, Technical University of regimes for a single-mode semiconductor laser with external feed-
back,” IEEE J . Quantum Electron., vol. 24, pp. 1242-1247, July
Denmark, and D. Salech for helpful discussions. 1988.
J. M@rkand B. Tromborg, “The mechanism of mode selection for
an external cavity laser,” IEEE Photon. Technol. Lett., vol. 2, pp.
REFERENCES 21-23,Jan. 1990.
R. Ries and F. Sporleder, “Low frequency instabilities of laser diodes
[l] L. Goldberg, H. F. Taylor, A. Dandridge, J. F. Weller, and R. 0. with optical feedback,” in Proc., 8th ECOC, Cannes, France, Sept.
Miles, “Spectral characteristics of semiconductor lasers with optical 1982, pp. 285-290.
feedback,” IEEE J. Quantum Electron., vol. QE-18, pp. 555-564, N. Schunk and K. Petermann, “Measured feedback-induced intensity
Apr. 1982. noise for 1.3 pm DFB laser diodes,” Electron. Lett., vol. 25, pp.
[2] D. Lenstra, B. H. Verbeek, and A . J. den Boef, “Coherence collapse 63-64, Jan. 1989.
in single-mode semiconductor lasers due to optical feedback,” IEEE J. Mdrk, M. Semkow, and B. Tromborg, “Measurement and theory
J . Quantum Electron., vol. QE-21, pp. 674-679, June 1985. of mode hopping in external cavity lasers,” Electron. Left., vol. 26,
[3] R. W. Tkach and A. R. Chraplyvy, “Regimes of feedback effects in .. 609-610.,Am.
DD. . 1990.
1.5 pm distributed feedback lasers,” J. Lightwave Technol., vol. [28] D. Lenstra, “Statistical theory of the multistable external-feedback
LT-4, pp. 1655-1661, NOV. 1986. laser,” Opt. Commun., vol. 81, pp. 209-214, Feb. 1991.
[4] H. Olesen, J. H. Osmundsen, and B. Tromborg, “Nonlinear dynam- [29] J. S. Cohen and D. Lenstra, “Spectral properties of the coherence
ics and spectral behavior for an external cavity laser,” IEEE J . Quan- collapsed state of a semiconductor laser with delayed optical feed-
tum Electron., vol. QE-22, pp. 762-773, June 1986. back,” IEEE J . Quantum Electron., vol. 25, pp. 1143-1151, June
[5] S . L. Woodward, T. L. Koch, and U. Koren, “The onset of coher- 1989.
ence collapse in DBR lasers,” IEEE Photon. Technol. Lett., vol. 2, [30] J. S. Cohen, F. Wittgrefe, M. D. Hoogerland, and J. P. Woerdman,
pp. ‘391-394, June 1990. “Optical spectra of a semiconductor laser with incoherent optica:
[6] Y. Cho and T. Umeda, “Observation of chaos in a semiconductor feedback,” IEEE J. Quantum Electron., vol. 26, pp. 982-990, June
laser with delayed feedback,” Opt. Commun.,vol. 59, pp. 131-136, 1990.
Aug. 1986. [31] B. Dorizzi, B. Grammaticos, M. Le Berre, Y. Pomeau, E. Ressayre,
[7] J. M&k, B. Tromborg, and P. L. Christiansen, “Bistability and low- and A. Tallet, “Statistics and dimension of chaos in differential delay
frequency fluctuations in semiconductor lasers with optical feedback: systems,” Phys. Rev. A, vol. 35, pp. 328-339, Jan. 1987.
A theoretical analysis,” ZEEE J . Quantum Electron., vol. 24, pp. [32] J. Wang and K. Petermann, “Noise analysis of semiconductor lasers
123-133, Feb. 1988. within the coherence collapse regime,” IEEE J . Quantum Electron.,
[8] J. M@rkand K. Kikuchi, “Low-frequency fluctuations and chaos in vol. 27, pp. 3-9, Jan. 1991.
a distributed-feedback semiconductor laser with optical feedback,” [33] H. Temkin, N. A. Olsson, J. H. Abeles, R. A. Logan, and M. B.
in Tech. Dig., Conf. Lasers and Electro-Optics, Anaheim, CA, Apr. Panish, “Reflection noise in index-guided InGaAsP lasers,” IEEE J .
1988, paper WM-14, pp. 216-217. Quantum Electron., vol. QE-22, pp. 286-293, Feb. 1986.
[9] G. C. Dente, P. S. Durkin, K. A. Wilson, andC. E. Moeller, “Chaos [34] C. H. Henry, and R. F. Kazarinov, “Instability of semiconductor
in the coherence collapse of semiconductor lasers,” ZEEE J . Quantum lasers due to optical feedback from distant reflectors,” IEEEJ. Quan-
Electron., vol. 24, pp. 2441-2447, Dec. 1988. tum Electron., vol. QE-22, pp. 294-301, Feb. 1986.
[IO] J. Sacher, W. Elsasser, and E. 0. Gobel, “Intermittency in the co- [35] P. BergC, Y. Pomeau, and C. Vidal, Order within Chaos. New
herence collapse of a semiconductor laser with external feedback,” York: Wiley, 1984.
Phys. Rev. Lett., vol. 63, pp. 2224-2227, Nov. 1989. [36] M. Fujiwara, K. Kubota, and R. Lang, “Low-frequency intensity
[ l l ] B. Tromborg and J. M$rk, “Nonlinear injection locking dynamics fluctuation in laser diodes with external optical feedback,” Appl.
and the onset of coherence collapse in external cavity lasers,” IEEE Phys. Lett., vol. 38, pp. 217-220, Feb. 1981.
J . Quantum Electron., vol. 26, pp. 642-654, Apr. 1990. [37] K. Tatah and E. Garmire, “Low-frequency intensity noise resonance
[12] -, “Stability analysis and the route to chaos for laser diodes with in an external cavity GaAs laser for possible laser characterization,”
optical feedback,” IEEE Photon. Technol. Lett., vol. 2, pp. 549- IEEE J. Quantum Electron., vol. 25, pp. 1800-1807, Aug. 1989.
552, Aug. 1990. [38] M. Port and K. J. Ebeling, “Intensity noise dependence on the injec-
[13] J. M$rk, J. Mark, and B. Tromborg, “Route to chaos and competi- tion current of laser diodes with optical feedback,” IEEEJ. Quantum
tion between relaxation oscillations for a semiconductor laser with Electron., vol. 26, pp. 449-455, Mar. 1990.
optical feedback,” Phys. Rev. Lett., vol. 65, pp. 1999-2002, Oct. [39] J. D. Park, D. S . Seo, and J. G. McInerney, “Self-pulsations in
1990. strongly coupled asymmetric external cavity semiconductor lasers,”
[14] F. Favre, D. Le Guen, and J. C. Simon, “Optical feedback effects IEEE J . Quantum Electron., vol. 26, pp. 1353-1362, Aug. 1990.
upon laser diode oscillation field spectrum,” IEEE J . Quantum Elec- [40] J. Sacher, W. Elsasser, and E. 0. Gobel, “Nonlinear dynamics of
tron., vol. QE-18, pp. 1712-1717, Oct. 1982. semiconductor laser emission under variable feedback conditions,”
[15] R. Wyatt and W. J. Devlin, “10 kHz linewidth 1.5 pm InGaAsP IEEE J . Quantum Electron., vol. 27, pp. 373-379, Mar. 1991.
external cavity laser with 55 nm tuning range,” Electron. Lett., vol. [41] J. Sacher and W. Elsasser, private communication.
19, pp. 110-112, Feb. 1983. [42] T. Mukai and K. Otsuka, “New route to optical chaos: Successive-
[16] J. Mark, E. BZdtker, and B. Tromborg, “Measurement of Rayleigh subharmonic oscillation cascade in a semiconductor laser coupled to
I08 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 28, NO. I, JANUARY 1992

an external cavity,” Phys. Rev. Lett., vol. 55, pp. 1711-1714, Oct. Jesper M#rk was born in Copenhagen, Denmark,
1985. on September 2, 1962. He received the M.Sc. and
[43] H. G. Winful, Y. C. Chen, and J. M. Liu, “Frequency locking, Ph.D. degrees in electrical engineering from the
quasiperiodicity, and chaos in modulated self-pulsing semiconductor Technical University of Denmark, Lyngby, in
lasers,” Appl. Phys. Lett., vol. 48, pp. 616-618, Mar. 1986. 1986 and 1988, respectively.
[44] D. Baums, W. ElsaBer, and E. 0. Gobel, “Farey tree and devil’s Since 1988 he has been with TFL Telecommu-
staircase of a modulated external-cavity semiconductor laser, ” Phys. nications Research Laboratory, HBrsholm, Den-
Rev. Lett, vol. 63, pp. 155-158, July 1989. mark. His main research interests are optical feed-
[45] S. Schuster, T. Wicht, and H. Haug, “Theory of the dynamical re- back effects and ultrafast dynamics in
laxation oscillations and of frequency locking in a synchronously optoelectronic devices.
pumped laser diode,” IEEE J. Quantum Electron., vol. 27, pp. 205-
211, Feb. 1991.
[46] G. P. Agrawal and N. K. Dutta, Long-Wavelength Semiconductor
Lasers. New York: Van Nostrand Reinhold, 1986.
[47] B. Tromborg, H. Olesen, X. Pan, and S. Saito, “Transmission line Bjarne Tromborg was born in Denmark in 1940.
description of optical feedback and injection locking for Fabry-Perot He received the M.Sc. degree in physics and
and DFB lasers,” IEEEJ. Quantum Electron., vol. QE-23, pp. 1875- mathematics from the Niels Bohr Institute, Co-
1889, Nov. 1987. penhagen, Denmark, in 1968.
[48] B. Tromborg, J. H. Osmundsen, and H. Olesen, “Stability analysis From 1986 to 1977 he was a Research Associ-
for a semiconductor laser in an external cavity,” IEEE J. Quantum ate at NORDITA and the Niels Bohr Institute. His
Electron., vol. QE-20, pp. 1023-1032, Sept. 1984. research field was theoretical elementary particle
[49] J. Mdrk, “Nonlinear dynamics and stochastic behavior of semicon- physics, in particular, analytic S-matrix theory and
ductor lasers with optical feedback,” Ph.D. dissertation, Laboratory electromagnetic corrections to hadron scattenng.
of Applied Mathematical Physics, Technical University of Denmark, Dunng this time he coauthored a research mono-
1988. graph on dispersion theory. From 1977 to 1979 he
[50] J. Helms and K. Petermann, “A simple analytical expression for the taught at a high school. Since 1979 he has been with TFL Telecommuni-
stable operation range of laser diodes with optical feedback,” IEEE cations Research Laboratory, H@holm, Denmark, and since 1987, as Head
J . Quantum Electron., vol. 26, pp. 833-836, May 1990. of the Optical Communications Group. Since 1991 he has been with the
[51] N. Schunk and K. Petermann, “Stability analysis for laser diodes University of Copenhagen as an Adjunct Professor of physics. His present
with short external cavities,” IEEE Photon. Technol. Lett., vol. 1, research interests in(Aude physics and technology of optoelectronic de-
pp. 49-51, Mar. 1989. vices.
G. A. Acket, D. Lenstra, A. J. den Boef, and B. H. Verbeek, “The Mr. Tromborg received the Electroprize from the Danish Society of En-
influence of feedback intensity on longitudinal mode properties and gineers in 1981.
optical noise in index-guided semiconductor lasers,” IEEE J. Quan-
tum Electron., vol. QE-20, pp. 1163-1169, Oct. 1984.
J. 0. Binder and G. D. Cormack, “Mode selection and stability of a
semiconductor laser with weak optical feedback,” IEEE J. Quantum
Electron., vol. 25, pp. 2255-2259, Nov. 1989. Jannik Mark was born in Denmark in 1957. He
J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical received the M.Sc. and Ph.D. degrees in electri-
Systems, and Bifurcations of Vector Fields. New York; Springer- cal engineering from the Technical University of
Verlag, 1983. Denmark, Lyngby, in 1983 and 1987, respec-
C. Grebogi, E. Ott, and J. A. Yorke, “Are three-frequency quasi- tively.
periodic orbits to be expected in typical nonlinear dynamical sys- Since 1986 he has been with TFL Telecommu-
tems?” Phys. Rev. Lett., vol. 51, pp. 339-342, Aug. 1983. nications Research Laboratory, Hgrsholm, Den-
J. S. Cohen, R. R. Drenten, and B. H. Verbeek, “The effect of op- mark. His present research interests are femto-
tical feedback on the relaxation oscillation in semiconductor lasers,” second pulse generation, ultrafast dynamics, and
IEEEJ. Quantum Electron., vol. 24, pp. 1989-1995, Oct. 1988. photonic switching.

You might also like