You are on page 1of 5

Journal of the Korean Physical Society, Vol. 65, No. 12, December 2014, pp.

2077∼2081

Photoluminescence Saturation and Exciton Decay Dynamics in Transition


Metal Dichalcogenide Monolayers

Min Ju Shin, Dong Hak Kim and D. Lim∗


Department of Applied Physics, College of Applied Science, Kyung Hee University, Yongin 449-701, Korea

(Received 17 September 2014, in final form 13 October 2014)

We report a photoluminescence (PL) and transient reflection spectroscopy study of the exci-
ton dynamics in monolayer transition-metal dichalcogenides (TMDs). PL saturation in monolayer
MoSe2 occurs at an excitation intensity more than two orders of magnitude lower than in monolayer
MoS2 . Transient reflection shows that nonlinear exciton-exciton annihilation is the dominant exci-
ton decay process in monolayer MoSe2 in contrast to the previously-reported linear exciton decay
in monolayer MoS2 . In addition, the exciton lifetime in MoSe2 , which is greater than 125 ps, is
more than an order of magnitude longer than the several-ps exciton lifetime in MoS2 . We find that
the dramatically different exciton decay mechanisms and PL saturation behaviors of the MoSe2 and
the MoS2 monolayers can be explained in terms of the difference in their exciton lifetimes.

PACS numbers: 73.21.Ac, 73.50.Gr, 78.47.+p, 78.55.HX


Keywords: Exciton dynamics, Photoluminescence, Transient reflection spectroscopy, MoS2 , MoSe2
DOI: 10.3938/jkps.65.2077

I. INTRODUCTION licity [13,14]. Based on the circular polarization of pho-


toluminescence (PL), the hole valley-spin lifetime greater
than 1 ns was proposed. Afterwards, the valley polariza-
The transition-metal dichalcogenide (TMD) MX2 , tion, as well as the valley coherence, was reported for
such as MoS2 , MoSe2 , WS2 and WSe2 , has a layered other monolayer and bilayer TMDs, such as WS2 and
structure, where each layer is bonded by a weak van der WSe2 [15, 16]. The exciton and the valley dynamics in
Waals interaction [1]. Although bulk MX2 is an indi- monolayer MoS2 have also been studied recently by us-
rect bandgap semiconductor, it exhibits a crossover from ing helicity-resolved transient absorption [17, 18], time-
an indirect to a direct bandgap in the monolayer limit resolved photoluminescence [19], and Kerr rotation [20].
[2–4]. Since the discovery of the direct bandgap struc- The obtained exciton lifetimes of monolayer MoS2 are
ture in monolayer MoS2 , extensive research efforts have typically several picoseconds [18,19], and the valley depo-
been made on few-layer TMDs because of their excellent larization times range from sub-ps to several picoseconds
optical and electrical properties [2,5–7]. In addition, elec- [17,20], in stark contrast with the greater than 1 ns valley
trons in few-layer MX2 possess a valley degree of freedom lifetime estimated based on circular polarization of the
as in graphene [8–10]. Graphene has been the material PL [14]. On the other hand, a recent work on monolayer
used in valleytronics research, but its drawbacks, such MoSe2 showed a strongly density dependent exciton de-
as the inherent lack of inversion symmetry and bandgap cay and a relatively longer exciton lifetime [21]. Because
limit its applications [8,9]. In contrast, monolayer MX2 a long exciton lifetime is a prerequisite to valley-based
has an inherently broken inversion symmetry and direct devices, a better understanding of the exciton dynam-
bandgap structure at degenerate valleys, K and K’, lo- ics in atomically-thin TMDs is required for their future
cated at the corners of the hexagonal Brillouin zone [10, applications.
11]. Near the conduction and the valence band edges, In this paper, we report our work on the exciton de-
the electronic states are of a d-orbital character. There- cay dynamics in monolayer MoSe2 and MoS2 studied by
fore, the strong spin-orbit interaction in the d-orbital, using excitation intensity/fluence dependent photolumi-
together with the broken inversion symmetry, spin-splits nescence and transient reflection spectroscopy. We find
the valence band significantly and couples the spin and dramatically different PL saturation behaviors and exci-
the valley degree of freedoms in the monolayer MX2 [12]. ton decay mechanisms between MoSe2 and MoS2 mono-
The first optical generation of valley polarization was layers. We find that the difference in the exciton lifetimes
reported in 2012 for monolayer MoS2 by using optical he- can explain our results.

∗ E-mail: dlim@khu.ac.kr
-2077-
-2078- Journal of the Korean Physical Society, Vol. 65, No. 12, December 2014

II. EXPERIMENTS

MoS2 and MoSe2 monolayer samples are obtained by


using mechanical exfoliation from bulk crystals on silicon
substrates with a 280 nm-thick thermal SiO2 or a 70-
nm thick pulsed-laser-deposited Al2 O3 overlayer. The
monolayer is first identified by using a optical microscope
and then the confirmed by raman and photoluminescence
measurements.
A Ti:sapphire femtosecond oscillator with a sub-100
fs pulse duration and an 82 MHz repetition rate was
used to excite the monolayer MoSe2 samples, both in
the photoluminescence and the transient reflection spec-
troscopy measurements. The 632-nm line from a He-
Ne laser was used for MoS2 excitation. A long-working- Fig. 1. (Color online) Schematic diagram of the transient
distance 50X objective lens was used to focus the light reflection measurement setup.
onto the monolayers with a spot diameter of ∼1.4 m.
The PL measurement of MoSe2 was performed by using
either a 760-nm CW or a mode-locked pulse laser beam. as:
After passing through the objective lens, the pulse from
the Ti:sapphire oscillator was broadened to ∼250 ps due dN N γ
=g− − N 2, (1)
to group velocity dispersion. Band pass filters were used dt τ 2
to clean the laser beam before it was focused onto sam-
where g is the exciton generation rate, τ1 (τ ) is the linear
ples and edge pass filters were used to cut the laser line
exciton decay rate (exciton lifetime), and γ is the nonlin-
from PL emission. Figure 1 shows a schematic diagram of
ear exciton-exciton annihilation (EEA) coefficient. Be-
the transient reflection measurement setup. For the tran-
cause the steady-state exciton density N is constant for
sient reflection measurement, mode-locked 780-nm pulses
CW excitation, it can be expressed
 as
a function of exci-
from the oscillator were divided into pump and probe by 1 I
using a beam-splitter, and the pump-excitation-induced tation intensity I as N = τ r −1 + 1 + I0 . Here, I0
reflection change in the probe at the chopper frequency E
= 2τ 2pγa (Ep : excitation photon energy, a: absorbance
was measured as a function of the pump-probe delay by of the sample) is the excitation intensity at which the
using a lock-in amplifier. Here, a combination of a half- dominant decay mechanism changes from linear exciton
wave plate and an analyzer was used to block the pump decay to nonlinear EEA.
before the detector. We fit the PL response from monolayer MoSe2 and plot
the result in Fig. 2(a) as a red line. In the low excitation
limit (I  I0 ), the nonlinear exciton-exciton interaction
III. RESULTS AND DISCUSSION is insignificant, and the PL intensity is proportional to
the excitation intensity. If the excitation intensity is high
(I  I0 ), EEA dominates the decay process, and the PL
Figure 1 shows the PL intensity as a function of the intensity becomes a nonlinear function of the excitation
excitation laser intensity for monolayer MoSe2 and MoS2 intensity. The contributions of the linear and the non-
samples. The PL responses are dramatically different for linear decay terms to the total decay rate are calculated
the MoSe2 and the MoS2 monolayers. The PL from the and plotted in the inset of Fig. 2(a). The decay rate due
MoS2 monolayer on both SiO2 and Al2 O3 overlayers is to EEA begins to exceed that due to linear exciton decay
almost linear throughout the entire range of excitation at a very low excitation intensity, ∼0.5 μW/μm2 . In con-
intensity (MoS2 monolayer on SiO2 displays a slightly trast, the PL from monolayer MoS2 doesn’t show much
more nonlinear behavior, probably due to the low dielec- deviation from a linear response even at a more than
tric constant of SiO2 ). The PL from MoSe2 , however, is two orders of magnitude higher excitation intensity, ∼80
completely different, showing a highly nonlinear behav- μW/μm2 .
ior, even at very low excitation intensity. Because the We also perform PL measurements by using femtosec-
PL intensity is proportional to the exciton density, the ond pulse excitation, and the resulting integrated PL
nonlinearity means that the steady-state exciton density from monolayer MoSe2 is plotted in Fig. 2(b) as a func-
in monolayer MoSe2 is not proportional to the excitation tion of the excitation pulse’s fluence. The integrated PL
laser intensity (the nonlinearity is not from saturation of saturates much faster than it does in CW excitation case
the absorption, as will be shown later) and that the exci- at the same laser power. Because the exciton density is
ton decay occurs via a strong exciton-exciton interaction. very high for pulse excitation case, especially just after
The rate equation for exciton density N can be written excitation, the excitons will decay mostly through EEA.
Photoluminescence Saturation and Exciton Decay Dynamics · · · – Min Ju Shin et al. -2079-

Fig. 2. (Color online) PL intensity vs. excitation inten-


sity graphs for monolayer MoS2 on 280-nm-thick SiO2 /Si (red
solid circles) and 70-nm-thick Al2 O3 /Si (open black circles),
and for monolayer MoSe2 on 280-nm-thick SiO2 /Si (blue solid
circles).

In this region, the integrated PL increases as ln(N0 ) as


the injected exciton density N0 is increased. If no ap-
preciable saturation of absorption is asuumed, the inte-
grated PL then has a functional form of ln(F) for a pulse
fluence F. We can see that the PL saturation in the high
fluence region can be fitted well by using this functional
form, supporting again the presence of a strong EEA
mechanism in monolayer MoSe2 .
To understand the physics underlying the dramatically
Fig. 3. (Color online) (a) PL intensity vs. excitation in-
different PL saturation in monolayer TMDs, we per- tensity of monolayer MoSe2 . The red line is a fitting result.
form degenerate transient reflection spectroscopy mea- (b) Integrated PL vs. excitation pulse fluence of monolayer
surements on monolayer MoSe2 and study its exciton MoSe2 . The red line is a fitting result. The excitation is
dynamics in the time domain. Figure 3 displays a typical made using a CW and a mode-locked 760-nm beam from a
differential reflection (DR) signal for monolayer MoSe2 Ti:sapphire oscillator.
measured by using degenerate pump-probe spectroscopy
at a wavelength of 780 nm. During the overlap of the
pump and the probe pulses, a strong coherent artifact at first and then slows down as the probe delay is in-
signal shows up, making it hard to observe the build-up creased, as expected from the density-dependent EEA
process of excitons. Still, we can ignore the exciton de- rate. This is different from monolayer MoS2 , where
cay during the pump-probe overlap because the exciton only a multi-exponential, linear exciton decay process
lifetime is much longer than the pulse duration. In the has been reported. If EEA dominates over linear exciton
inset of Fig. 3, the maximum DR signal is plotted as a decay, the linear exciton decay term in Eq. (1) can be
function of pump fluence. The plotted maximum DR is ignored, and the exciton density then evolves with time
not a linear function of the pump fluence. We should as follows [21]:
point out that although the absorption may seem to be
saturated, it is not. The maximum DR signal of the N0
− 1 = γN0 t, (2)
probe is less than 1%, so the change in the absorption N (t)
during the pump excitation should be small, too. There-
fore, the nonlinear dependence of the DR signal should Equation (2) indicates that if EEA dominates over lin-
be related to other causes, such as bandgap renormal- ear exciton decay, the slope of the NN(t)
0
− 1 graph should
ization and/or a pump-heating induced A-exciton peak be proportional to the injected exciton density and, thus,
shift. Irrespective of its origin, we can exclude saturation to the pump fluence. In a linear exciton decay process,
of absorption and safely assume that the injected exciton however, the slope is given by the inverse of the life-
density is proportional to the pump fluence. time time, τ −1 , which has no direct correlation with
After pump excitation, the DR signal relaxes rapidly the pump fluence. In Fig. 4(a), NN(t) 0
− 1 is plotted as
-2080- Journal of the Korean Physical Society, Vol. 65, No. 12, December 2014

Fig. 4. Differential reflection of monolayer MoSe2 . The


center wavelength of the pump and the probe is 780 nm and
the pump fluence is 53 μJ/cm2 . (Inset) Maximum DR as a
function of the excitation fluence.

a function of the probe delay for several pump fluencies.


Clearly, the slope of the graphs increases as the pump
fluence is increased, supporting a strong EEA in mono-
layer MoSe2 . A similar density-dependent decay rate has
been reported in a previous report [21]. The slope, how-
ever, is not exactly proportional to the injected exciton
density. This deviation may be explained by the resid-
ual linear decay term. Although the linear exciton decay
term is ignored in Eq. (2), it always contributes to the
Fig. 5. (Color online) (a) Plot of NN(t)
0
− 1 vs. probe delay
total excition decay rate. This is increasingly so as the
pump fluence is decreased. It adds an additional slope for pump fluencies of 105, 53, 26 μJ/cm2 (from top to bot-
to Eq. (2) and causes the total slope to deviate from the tom). The DR is assumed to be proportional to the exciton
density. The red lines are linear fits to the initial part (0 ∼ 40
linear dependence, especially at low pump fluences.
ps delay) of the graphs. (b) Exponential fitting to the latter
When the exciton density is low, its linear decay can part (50 − 300 ps) of DR at a pump fluence of ∼5.3 μJ/cm2 .
dominate over the nonlinear EEA. This condition is met
at ‘low pump fluence’ and ‘long delay time’. We can thus
extract the exciton lifetime, or its lower limit because
EEA always co-exists, by fitting the long-delay part of ing this dramatic change. The relative magnitude of the
the DR data pumped at a low fluence. Figure 4(b) dis- nonlinear EEA compared to the linear exciton decay is
plays the DR signal measured at the lowest excitation Nγτ /2. At a given exciton density, the longer the exci-
fluence, ∼5.3 J/cm2 , together with a single-exponential ton lifetime or the larger the EEA coefficient is, the more
fitting result as a red line (fitting region: delay > 50 ps). significant the EEA becomes. As can be seen in the tran-
The exciton lifetime obtained is around ∼125 ps. This is sient reflection for monolayer MoSe2 , the exciton-exciton
more than an order of magnitude longer than the exciton interaction is strong in 2-D TMDs. However, the EEA
lifetime of monolayer MoS2 , which has been reported to coefficients of MoS2 and MoSe2 monolayers should be of
be several picoseconds [18, 19]. Why their exciton life- the same order, because the exciton binding energies and,
times are so different is not clear. A 2D TMD mono- thus, the exciton radii are similar. Therefore, this does
layer has a more complicated structure than graphene not explain the difference in the decay mechanism. On
and, thus, has much more various point defects and dis- the other hand, the measured exciton lifetime of mono-
location cores [22]. One possible reason may be that S- layer MoSe2 is more than an order of magnitude longer
related point defects such as vacancies are formed more than that of monolayer MoS2 . Therefore, the linear de-
easily in MoS2 than Se-related defects in MoSe2 during cay rate is much smaller in monolayer MoSe2 than in
mechanical exfoliation. monolayer MoSe2 , leaving EEA as the dominant decay
Our work shows that the PL saturation behavior and mechanism.
the exciton decay mechanism can change from one TMD The dramatically different PL saturation behaviors
to another. Here, we will discuss the physics underly- can be explained also by the exciton lifetime. The
Photoluminescence Saturation and Exciton Decay Dynamics · · · – Min Ju Shin et al. -2081-

crossover excitation intensity from EEA to linear exci- [4] S. Tongay, J. Zhou, C. Ataca, K. Lo, T. S. Matthews,
E J. Li, J. C. Grossman and J. Wu, Nano Lett. 12, 5576
ton decay scales as I0 = 2τ 2pa . The absorbance a is re-
ported to be around ∼0.05 for both MoS2 and MoSe2 (2012).
monolayers [23]. Therefore, the main difference is the [5] R. S. Sundaram, M. Engel, A. Lombardo, R. Krupke, A.
C. Ferrari, Ph. Avouris and M. Steiner, Nano Lett. 13,
exciton lifetime τ . Because I0 is proportional to τ −2 ,
1416 (2013).
the longer exciton lifetime in monolayer MoSe2 causes a [6] Y. Yoon, K. Ganapathi and S. Salahuddin, Nano Lett.
crossover from linear to nonlinear exciton decay to occur 11, 3768 (2011).
at a more-than two orders of magnitude lower excitation [7] B. Radisavljevic, A. Radenovic, J.Brivio, V. Giacometti
intensity than in monolayer MoS2 , as observed in our and A. Kis, Nat. Nanotechnol. 6, 147 (2011).
experiment. [8] D. Xiao, W. Yao and Q. Niu, Phys. Rev. Lett. 99, 236809
(2007).
[9] W. Yao, D. Xiao and Q. Niu, Phys. Rev. B 77, 235406
(2008).
IV. CONCLUSION [10] D. Xiao, G.-B. Liu, W. Feng, X. Xu and W. Yao, Phys.
Rev. Lett. 108, 196802 (2012).
We study the exciton decay dynamics in monolayer [11] T. Cao et al., Nat. Commun. 3, 887 (2012).
TMDs. We observe a highly nonlinear PL response in [12] Z. Y. Zhu, Y. C. Cheng and U. Schwingenschlogl, Phys.
Rev. B 84, 153402 (2011).
monolayer MoSe2 , in contrast to the almost linear one in
[13] H. Zeng, J. Dai, W. Yao, D. Xiao and X. Cui, Nat. Nan-
monolayer MoS2 . Transient differential reflection shows otechnol. 7, 490 (2012).
that the exciton decay in monolayer MoSe2 is dominated [14] K. F. Mak, K. He, J. Shan and T. F. Heinz, Nat. Nan-
by exciton-exciton annihilation, in contrast to the linear otechnol. 7, 494 (2012).
exciton decay in monolayer MoS2 . We find that the dif- [15] A. M. Jones et al., Nat. Nanotechnol. 11, 1 (2013).
ferences in the exciton decay mechanism and the PL sat- [16] B. Zhu, H. Zeng, J. Dai, Z. Gong and X. Cui, PNAS
uration behavior in monolayer TMDs can be explained 116, 11606 (2014).
in terms their different exciton lifetimes. [17] T. Korn, S. Heydrich, M. Hirmer, J. Schmutzler and C.
Schueller, Appl. Phys. Lett. 99, 102109 (2011).
[18] Q. Wang, S. Ge, X. Li, J. Qiu, Y. Ji, J. Feng and D. Sun,
ACS Nano 7, 11087 (2013).
ACKNOWLEDGMENTS [19] D. Lagarde, L. Bouet, X. Marie, C. R. Zhu, B. L. Liu,
T. Amand, P. H. Tan and B. Urbaszek, Phys. Rev. Lett.
112, 047401 (2014).
This work was supported by grants from the Kyung
[20] C. R. Zhu, K. Zhang, M. Glazov, B. Urbaszek,
Hee University Research Fund (KHU- 20110464). T. Amand, Z. W. Ji, B. L. Liu and X. Marie,
[arXiv:1407.5862v1] [cond-mat.mes-hall] (2014).
[21] N. Kumar, Q. Cui, F. Ceballos, D. He, Y. Wang and H.
REFERENCES Zhao, Phys. Rev. B 89, 125427 (2014).
[22] W. Zhou, X. Zou, S. Najmaei, Z. Liu, Y. Shi, J. Kong,
J. Lou, P. M. Ajayan, B. I. Yakobson and J.-C. Idrobo,
[1] K. S. Novoselov, D. Jiang, F. Schedin, T. J. Booth, V. Nano Lett. 13, 2615 (2013).
V. Khotkevich, S. V. Morozov and A. K. Geim, Proc. [23] M. Bernardi, M. Palummo and J. C. Grossman, Nano
Natl Acad. Sci. USA 102, 10451 (2005). Lett. 13, 3664 (2013).
[2] A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C. Y.
Chim, G. Galli and F. Wang, Nano Lett. 10, 1271 (2010).
[3] K. F. Mak, C. Lee, J. Hone, J. Shan and T. F. Heinz,
Phys. Rev. Lett. 105, 136805 (2010).

You might also like