Article
https://doi.org/10.1038/s41467-023-37481-1
All-optical control of high-purity trions in
nanoscale waveguide
Hyeongwoo Lee
1
, Yeonjeong Koo
1
, Shailabh Kumar
2,3
, Yunjo Jeong
4
,
Dong Gwon Heo
5
, Soo Ho Choi
6
,HuitaeJoo
1
,MinguKang
1
,
Radwanul Hasan Siddique
2,3
,KiKangKim
6,7
, Hong Seok Lee
5
,SangminAn
5
,
Hyuck Choo
2,8
&Kyoung-DuckPark
1
The generation of high-purity localized trions, dynamic exciton
–
trion inter-
conversion, and their spatial modula
tion in two-dimensional (2D) semi-
conductors are building blocks for
the realization of trion-based
optoelectronic devices. Here, we prese
nt a method for the all-optical control
of the exciton-to-trion conversion process and its spatial distributions in a
MoS
2
monolayer. We induce a nanoscale str
ain gradient in a 2D crystal trans-
ferred on a lateral metal
–
insulator
–
metal (MIM) waveguide and exploit pro-
pagating surface plasmon polaritons (S
PPs) to localize hot electrons. These
signi
fi
cantly increase the electrons and ef
fi
ciently funnel excitons in the lateral
MIM waveguide, facilitating complete exciton-to-trion conversion even at
ambient conditions. Additionally, w
emodulatetheSPPmodeusingadaptive
wavefront shaping, enabling all-optica
l control of the exciton-to-trion con-
version rate and trion distribution in a
reversible manner. Our work provides a
platform for harnessing exci
tonic quasiparticles ef
fi
ciently in the form of
trions at ambient conditions, enabling high-ef
fi
ciency photoconversion.
The spatial control of excitonic quasiparticles in two-dimensional (2D)
semiconductors has been extensively studied for the development of
various exciton-based optoelectronic devices, especially facilitating
intermedium of the electronic system and optical system, as well as
highly ef
fi
cient light-harvesting devices
1
–
4
. The generation of drift-
induced exciton
fl
ux using various strain gradient geometries has been
widely adopted in manipulating the spatial distributions of excitonic
quasiparticles in transition metal dichalcogenide (TMD) monolayers
(MLs)
5
–
7
. However, because thermally driven exciton diffusion sig-
ni
fi
cantly reduces exciton
fl
ux, the funneling ef
fi
ciency of a neutral
exciton (X
0
) at room temperature can be very low
5
—
as low as < 3%,
according to a recent experimental study
6
.Meanwhile,withn-type
TMD MLs under a similar strain gradient geometry, the excess elec-
trons are funneled together with X
0
and converted to trions (X-) via an
exciton-to-trion conversion process. The ef
fi
ciency of this exciton-to-
trion conversion can reach 100% in a WS
2
ML suspended on a
microhole-based strain gradient, because strain-induced modi
fi
cation
of the bandgap increases the spatial overlap between X
0
and electrons
6
and affects Fermi level in the way of decreasing Schottky barrier
height
8
. Given the characteristics of X-, particularly the high-ef
fi
ciency
generation and reactivity to the external electric
fi
eld, the exciton-to-
trion conversion can be a promising alternative to the inef
fi
cient fun-
neling process of X
0
. In comparison with the large-area electrical
9
and
chemical doping
10
methods, exploiting strain gradient geometry
Received: 14 December 2022
Accepted: 20 March 2023
Check for updates
1
Department of Physics, Pohang University of Science and Technology (POSTECH), Pohang 37673, Republic of Korea.
2
Department of Medical Engineering,
California Institute of Technology (Caltech), Pasadena, CA 91125, USA.
3
Meta Vision Lab, Samsung Advanced Institute of Technology (SAIT), Pasadena, CA
91101, USA.
4
Institute of Advanced Composite Materials, Korea Institute of Science and Technology, Jeonbuk 55324, Republic of Korea.
5
Department of
Physics, Research Institute of Physics and Chemistry, Jeonbuk National University, Jeonju 54896, Republic of Korea.
6
Center for Integrated Nanostructure
Physics, Institute for Basic Science (IBS), Suwon 16419, Republic of Korea.
7
Department of Energy Science, Sungkyunkwan University (SKKU), Suwon 16419,
Republic of Korea.
8
Advanced Sensor Lab, Device Research Center, Samsung Advanced Institute of Technology (SAIT), Suwon 16678, Republic of Korea.
e-mail:
hslee1@jbnu.ac.kr
;
hyuck.choo@samsung.com
;
parklab@postech.ac.kr
Nature Communications
| (2023) 14:1891
1
1234567890():,;
1234567890():,;
facilitates higher conversion ef
fi
ciency, local injection of the electrons,
and ef
fi
cient spatial controllability, suitable to applications in nanos-
cale optoelectronic devices and trionic energy harvesting.
However, at ambient conditions, H
2
OandO
2
molecules physi-
sorbed onto the TMD ML surface signi
fi
cantly reduces electron
density
11
,
12
. Consequently, previous studies reported only minor por-
tions of X- in their radiative emissions compared to the dominant X
0
contributions, despite a 100% exciton-to-trion conversion ef
fi
ciency,
i.e., X
0
cannot be completely converted to X- at ambient conditions
because of a lack of electrons
6
,
13
. Moreover, exploiting mechanical
deformations to induce the proposed microscale 0D strain gradient
can be invasive in 2D crystals with limited durability. Moreover, it
restricts the direction of exciton
fl
ux and has a size mismatch with
nanoscale electronics in integrated circuits. Therefore, a noninvasive
and direction-controllable nanoscale platform with robust trion gen-
eration at ambient conditions is desirable for the practical application
of trionic devices.
Here, we present a versatile method for the all-optical control of
trion behavior in MoS
2
ML, including complete exciton-to-trion con-
version and localization, dynamic exciton
–
trion interconversion, and
spatial modulation of trions at ambient conditions. In our device, the
nanogap geometry of the lateral plasmonic metal
–
insulator
–
metal
(MIM) waveguide induces a 1D nanoscale strain gradient in the sus-
pended MoS
2
ML. The induced nanoscale strain gradient signi
fi
cantly
increases the funneling ef
fi
ciency, thus con
fi
ning X
0
to the nanogap
center
14
,
15
; however, as described earlier, there is a lack of electrons.
Therefore, the surface plasmon polariton (SPP) mode of the plasmonic
lateral MIM waveguide is utilized. The plasmon-induced hot electron
generation process enables the inj
ection of electrons from Au to the
MoS
2
ML
16
,
17
. These extra electrons are funneled toward the nanogap
center and locally increase the electron density, stimulating additional
exciton-to-trion conversion in the nanoscale region, i.e., the nanoscale
generation of radiative X- emission. Thus, we can either induce com-
plete conversion from X
0
to X- by activating the SPP mode or enable
dominant X
0
by deactivating the SPP mode, i.e., achieving polarization-
controllable exciton
–
trion interconversion between the dominant X
0
emission and high-purity X- emission.
Furthermore, we employ adaptive wavefront shaping using a
spatial light modulator (SLM) to dynamically manipulate the SPP mode
of the waveguide
18
,
19
, which in turn spatially modulates the exciton-to-
trion conversion region
20
.Weuseastepwisesequencefeedback
algorithm to enhance the plasmon intensity up to ~210% in the weak
SPP region and observe the corresponding dramatic increase in X-
emission intensity. Finally, we include the SPP effect in the drift-
diffusion model to investigate the quantitative localized electron
density under a nonhomogeneous strain pro
fi
le. By including the X-/X
0
ratio from experimental photoluminescence (PL) spectra and the mass
action model, we estimate an enhancement of ~10 times the localized
electron density for the full activation of the SPP mode.
Pre-characterization of al
l-optical trion control
platform
To achieve a complete exciton-to-trion conversion and all-optically
modulate their spatial distribution, we use a nanogap-based lateral
MIM waveguide device with adaptive excitation control, as shown in
Fig.
1
a. When the naturally n-doped MoS
2
ML is transferred to the
nanogap of the waveguide, the generated nanoscale strain gradient
funnels X
0
together with electrons, leading to the formation of X- at
the center of the nanogap (Supplementary Fig. 1). However, at ambient
conditions, the electron density of the MoS
2
ML noticeably decreases
owing to the presence of H
2
OandO
2
molecules physisorbed onto the
MoS
2
ML surface (Supplementary Fig. 2)
11
,
12
. Consequently, the number
of electrons is much smaller than the number of funneled X
0
at the
nanogap, resulting in incomplete e
xciton-to-trion conversion. By
contrast, the proposed nanogap-based lateral MIM waveguide device
with the designed SPP mode can supply extra electrons locally via
plasmon-induced hot electron generation, as illustrated in Fig.
1
b
(Supplementary Fig. 3)
16
,
17
. Hot electrons injected from the Au to the
MoS
2
ML in the SPP mode drift toward the nanogap center together
with X
0
,asshowninFig.
1
c, d (a detailed description of the physical
mechanism is presented using the theoretical model and experimental
data in Fig. 5). This additional provision of electrons in the SPP mode
results in a highly enhanced X- density and even leads to complete
exciton-to-trion conversion. The polarization-sensitive nature of the
SPP mode offers precise control of the exciton-to-trion conversion
ratio as a function of the SPP strength, and thus dynamic inter-
conversion between high-purity X- emission and dominant X
0
emis-
sion. Moreover, we adopt adaptive wavefront shaping to engineer the
SPP mode, which cannot be performed using conventional plasmonic
waveguides
18
,
19
.Figure
1
e depicts adaptive wavefront shaping, which
MoS
2
ML
SLM
Objective
lens
Exc.
Signal
5
Source
Detection
Optimal
wavefront
Initial
wavefront
Au
MoS
2
X
0
X-
e
e
e
h
h
Distance,
x
E
MoS
a
e
h
e
2
MoS
2
Wavefront shaping
Dynamic X- modulation
X
0
X-
Optimal phase mask
Initial phase mask
c
e
d
MoS
2
ML
Au
Au
Hot e
-
Funneling X
0
X- conversion
x
y
z
b
Adaptive optical
excitation
Fig. 1 | Illustration of propagating SPP and exciton-to-trion conversion with
dynamic optical modulation. a
Schematic diagram of all-optical trion control
platform operating with adaptive optical excitation. Green dashed line indicates
transferred MoS
2
ML.
b
Illustration of all-optical trion control platform facilitated
by nanoscale strain gradient, plasmon-induced hot electrons, and resultant
exciton-to-trion conversion.
c
Illustration of hot electron injection process from Au
to MoS
2
ML in SPP mode.
d
Strain (
ε
) and corresponding bandgap energy change
(
Δ
E
) diagram of MoS
2
ML as function of distance
x
, where gray region indicates
nanogap area.
e
Adaptive wavefront shaping using stepwise sequence feedback
algorithm to
fi
nd optimal phase mask (left) and dynamic excitonic emission
modulation through phase mask control (right).
Article
https://doi.org/10.1038/s41467-023-37481-1
Nature Communications
| (2023) 14:1891
2
uses a stepwise sequence feedback algorithm to optically modulate
the SPP mode of the waveguide. The optimal phase mask can sig-
ni
fi
cantly increase X- emission in the weak SPP region and enable
instantaneous switching between the dominant X
0
and X- emissions.
Polarization-dependent spatial distributions of
SPP, X
0
,andX-
We investigate the polarization-dependent activation of the SPP mode
in the waveguide and its effect on X
0
and X- densities
21
,
22
.Figure
2
a
shows the spatial distribution of the SPP with excitation polarization
across the waveguide, i.e., the waveguide is deactivated. As expected,
no evidence of the SPP mode is observed in the waveguide. However,
when the excitation polarization is along the waveguide, i.e., fully
activated, a strong SPP mode is observed in the waveguide, as shown in
Fig.
2
b. The spatial distribution of X
0
also exhibits a polarization-
dependent response. When the waveguide is deactivated, an enhanced
X
0
density is observed at the nanogap, which is attributed to the fun-
neling effect of the strain gradient of the nanogap, as shown in Fig.
2
c.
By contrast, when the waveguide is activated, the X
0
density near the
SPP mode gradually decreases, as shown in Fig.
2
d, e, where the
waveguide is partially activated and fully activated, respectively.
Interestingly, the spatial distribution of the X-/X
0
ratio exhibits an
opposite behavior from that of X
0
. When the waveguide is activated,
the gradual emergence of localized X- is observed in the SPP mode, as
shown in Fig.
2
f
–
h, where the waveguide is deactivated, partially acti-
vated, and fully activated, respectively. Speci
fi
cally, localization of the
X- emission is observed in the SPP mode, which is attributed to the 1D
strain gradient of the nanogap geometry and the SPP-induced local
enhancement of the electron density. The covariance map in Fig.
2
illustrates the resulting correlations of SPP, X
0
,andX-.Itcanbe
observed that X- is correlated with the SPP, whereas X
0
is antic-
orrelated with both the SPP and X-. This indicates a stepwise process
—
plasmon-induced hot electron generation, funneling of the injected
electrons toward the nanogap center, and additional exciton-to-trion
conversion. Note that we exclude the possible contribution from
defect-induced charges while con
fi
rming the role of lateral MIM
waveguide with control experiments at low excitation power (Sup-
plementary Figs. 6
–
8).
Radiative control of trions with complete exciton-
to-trion conversion
We then target the spot of the strong SPP mode and measure the time-
resolved photoluminescence (TRPL) traces, as shown in Fig.
3
a, b. The
TRPL traces are
fi
tted by a biexponential function with fast (
τ
1
)and
slow (
τ
2
)components
23
. Unlike previously reported plasmon-coupled
platforms, exhibiting signi
fi
cant decreases in decay time
24
–
26
, both
components derived from lateral MIM waveguide show minimal
changes in decay time compared to the ones from silicon. Speci
fi
cally,
the strain gradient geometry exploits the funneling of electrons and
high exciton-to-trion conversion ef
fi
ciency
6
,resultinginthesmaller
number of injected electrons to achieve complete exciton-to-trion
coversion. Therefore, we induce high electron density and corre-
spondingly enhanced trion emission while weakly coupled to the
plasmon, as shown in Fig.
3
b. Next, we measure spatial-dependent PL
responses with three different excitation polarizations, as shown in
Fig.
3
c. When the waveguide is deactivated, no spectral changes are
observed in the SPP mode. By contrast, when the waveguide is partially
activated, emergence of the X- emission is observed in the SPP mode,
although with signi
fi
cant emission of X
0
. Finally, when the waveguide is
fully activated, a high-purity X- emission with negligible X
0
emission is
produced. To further investigate the polarization-dependent behavior
of X
0
and X-, we
fi
t the PL spectra measured at the center of the SPP
mode to the Lorentz function, as shown in Fig.
3
d (Supplementary
Fig. 9). With the waveguide deactivated, a dominant X
0
emission is
observed with the X- shoulder, indicating low electron density at the
strain gradient center. We note that the intrinsic X- emission at the
deactivated waveguide originates from the funneling of the back-
ground electrons at the strain gradient and the intrinsic polarization
ratio of the excitation source (100:1). With the waveguide partially
activated, an additional SPP-mediated exciton-to-trion conversion is
Fig. 2 | Polarization-dependent hyperspectral imaging of SPP, X
0
,andX-at
nanoscale waveguide.
SPP images with excitation polarization across (
a
)and
along (
b
) waveguide.
c
–
e
X
0
PL images with different excitation polarizations.
f
–
h
PL images of X-/X
0
ratio with different excitation polarizations. Illustration of
scan area on the waveguide is based on Rayleigh scattering image (Supplementary
Figs. 4 and 5). Inset: (Anti-) correlation between SPP, X
0
, and X-.
Article
https://doi.org/10.1038/s41467-023-37481-1
Nature Communications
| (2023) 14:1891
3
observed, i.e., a decrease in the X
0
emission in contrast to an increase
in the X- emission. Finally, with the waveguide fully activated, a highly
dominant X- emission is produced, indicating that high-purity loca-
lized X- is achieved via a complete exciton-to-trion conversion (Sup-
plementary Figs. 10 and 11). The minor X
0
portion of the emission is
attributed to the diffraction-limited beam size (~450 nm), which
exceeds the nanogap size (~300 nm). These results indicate the
polarization-controllable exciton-to-trion conversion ratio and the
dynamic transition between the high-purity localized X- state and the
dominant X
0
state.
All-optical spatio-spectral modulation of
exciton
–
trion interconversion
Conventional static plasmonic waveguides have limitations in provid-
ing dynamic spatial controllability for the exciton-to-trion conversion
region owing to their
fi
xed SPP mode. To further enhance device
expandability, deterministic spatial control of the SPP mode is highly
desirable. To implement this, we use adaptive wavefront shaping with
the SLM, as illustrated in Fig.
4
a. We move the detection spot to the
weak SPP region and implement a sequence feedback algorithm, which
aims to maximize the target intensity by optimizing the wavefront
(Supplementary Figs. 12 and 13)
18
,
19
.Figure
4
b shows the evolution of
the plasmon intensity. The plasmon intensity gradually increases
during the wavefront shaping and consequently reaches an enhance-
ment of ~210% with the optimized phase mask, as shown in Fig.
4
c. This
trend implies that the SPP mode can be spatially modulated at the
desired location, enabling the spatial modulation of the exciton-to-
trion conversion region. We then compare the PL spectra with and
without the optimal phase mask, as shown in Fig.
4
d. Without the
optimal phase mask, the PL spectrum exhibits a dominant X
0
emission
because of a lack of electrons, as expected. By contrast, when the
optimal phase mask is used, X- emission becomes dominant, owing to
the provision of extra electrons by plasmon-induced hot electron
generation. We note that the increase in X- intensity is higher than the
decrease in X
0
intensity with the optimal phase mask. This is due to the
SPP-induced excitation of additional X
0
, consequently converted to X-
(Supplementary Figs. 10 and 11). This fully optical process offers non-
invasive modulation with excellent repeatability. This allows instanta-
neous exciton-to-trion conversions at desired locations, enabling
dynamic switching between the dominant X
0
and X- emissions, as
indicated by the two spectra in Fig.
4
d
–
f (Supplementary Figs. 14
and 15).
Theoretical investigation of plasmo-excitonic
transport and conversion dynamics
We analyze the drift-diffusion model using experimentally obtained
Kelvin probe force microscopy (KPFM) data to investigate the physical
mechanism of electron funneling and the related exciton-to-trion
conversion dynamics. The movement of electrons at the nanogap of
the waveguide can be experimentally estimated from the work func-
tion image (Supplementary Fig.
16). Note that work function
φ
=
E
vac
−
E
F
,where
E
vac
is the vacuum level and
E
F
is the Fermi level. As
shown in Fig.
5
a, b, an increase in the work function is observed in the
gradient region of tensile strain in the nanogap, which is in good
agreement with the results of a previous study
27
. Because the gradual
increase in the work function shown in Fig.
5
c reaches its maximum at
the nanogap center, the plasmon-induced hot electrons at the inter-
face of Au and MoS
2
ML in the SPP mode can be funneled into the
nanogap center. As a subsequent step, we theoretically estimate the
spatial distribution of the electron and X
0
in the presence of the
nanoscale strain gradient. First, we obtain the
fi
tted line-shape func-
tion based on the topography pro
fi
le of the MoS
2
ML suspended on the
nanogap, as shown in Fig.
5
d (Supplementary Fig. 17). The spatial dis-
tribution of the photoexcited excitons
n
(
x
) can be derived by solving
the steady-state continuity condition for the excitonic diffusion and
drift currents, as follows:
∇
ð
D
∇
n
ð
x
ÞÞ
+
∇
ð
μ
n
ð
x
Þ
∇
u
ð
x
ÞÞ
n
ð
x
Þ
τ
n
2
ð
x
Þ
R
A
+
S
ð
x
Þ
=0,
ð
1
Þ
where
D
∇
n
(
x
) is the excitonic diffusion current term,
μ
n
(
x
)
∇
u
(
x
)is
the excitonic drift current term
6
,
28
,
D
is the diffusion coef
fi
cient,
1.6
1.8
2.0
2.2
1
2
3
1
2
3
1
2
3
Fit
X
0
X-
1.7
1.8
1.9
0.0
0.5
1.0
1.5
1.7
1.8
1.9
1.7
1.8
1.9
Energy (eV)
Distance (μm)
Energy (eV)
Energy (eV)
Norm. PL intensity ( a. u.)
0
1
90
45
0
PL intensity ( a. u.)
Energy (eV)
0
45
90
0
45
90
MoS
2
a
c
d
Plasmon-induced
hot e
-
Funneling of e
-
and X
0
Pol. control
X- generation
Waveguide
Exc.
y
x
x
y
pump
04812
1
10
-2
1
10
-1
1
10
0
Si
Waveguide
Time (ns)
1
=0.26 ns
2
=2.98 ns
1
=0.24 ns
2
=2.87 ns
b
Norm. PL intensity ( a. u.)
Fig. 3 | Complete exciton-to-trion conversion. a
Illustration of exciton-to-trion
conversion process assisted by plasmon-induced hot electrons.
b
Normalized TRPL
traces of MoS
2
monolayer on silicon (black) and waveguide structure (red).
c
Spatial
dependent PL spectra obtained by vertically crossing waveguide (white dashed
line) with different excitation polarizations.
d
Corresponding PL spectra obtained
at center of SPP mode
fi
tted to Lorentz function.
Article
https://doi.org/10.1038/s41467-023-37481-1
Nature Communications
| (2023) 14:1891
4
u
(
x
)=
E
g
−
0.05
ε
(
x
) is the strain-induced bandgap (
E
g
) change under
the strain pro
fi
le
ε
(
x
)ofMoS
2
ML
29
,
μ
=
D
/
k
B
T
is the mobility (
k
B
is the
Boltzmann constant, and
T
is the temperature),
R
A
is the Auger
recombination rate,
τ
is the exciton lifetime, and
S
ð
x
Þ
=
I
0
2
πσ
2
e
x
2
=
2
σ
2
is
the exciton generation rate in a Gaussian illumination pro
fi
le (
I
0
is the
intensity and
σ
=FWHM
=
2
ffiffiffiffiffiffiffiffiffiffiffi
2ln2
p
). The diffusion coef
fi
cient values,
exciton lifetime, and Auger recombination rate of the MoS
2
ML are
obtained from previous studies
30
–
32
. We note that, at this stage, the
photoexcited exciton density
n
(
x
) includes all kinds of photoexcited
excitons, e.g., neutral excitons (X
0
) and charged excitons (X-). The
spatial distribution of the electron
n
e
(x) can be described based on an
assumption of the absence of electron generation by illumination (
S
(
x
)
=0,
τ
=0,and
R
A
=0).Thisyields
n
e
ð
x
Þ
=
N
0
e
∇
u
c
ð
x
Þ
=
k
B
T
R
e
∇
u
c
ð
x
Þ
=
k
B
T
xdx
,
ð
2
Þ
where
N
0
is the number of electrons in the entire area, and
∇
u
c
(
x
)isthe
strain-induced change in the conduction band
6
.Figure
5
eshowsthe
calculated pro
fi
les of the photoexcited exciton and electrons densities.
The experimental results indicate that the electron density increases at
the center of the nanogap, as expect
ed. Consequently, the electron
density decreases in the vicinity of the nanogap because of the
funneling of background electrons toward the nanogap center.
Meanwhile, with regard to the densi
ty of the photoexcited excitons,
a similar tendency is exhibited; th
e photoexcited exciton density is
funneled toward the nanogap center. Note that we subtract the
photoexcited exciton density obtained without the strain gradient
from the photoexcited exciton density with the strain gradient to
clearly demonstrate the effect of strain gradient and consequently
evaluate the density of drifted photoexcited excitons while excluding
the effect of optical excitation (Supplementary Fig. 18).
The term
N
0
R
e
∇
u
c
ð
r
Þ
=
k
B
T
xdx
in
n
e
(
x
) can be considered as an integration
constant, which is related to the global defect density of the sample
6
.If
we de
fi
ne this global defect density of the sample as
α
,thenan
increasing
α
indicates the provision of extra electrons, i.e.,
α
is pro-
portional to the electron density
n
e
(
x
). In our experiment, increasing
α
can be realized by plasmon-induced hot electrons, as it increases the
background electron density near the nanogap. Subsequently, we
gradually increase
α
and plot the evolution of the spatial distribution of
the electrons
n
e
(
x
) to investigate the electron density at the nanogap
center when plasmon-induced hot electron generation occurs in the
SPP mode. With increased background electron density, a signi
fi
cant
increase in the electron density is observed at the nanogap center, as
shown in Fig.
5
f. In this case, we now consider the contribution of X-
because the actual solution of drift-diffusion model
n
(
x
)isthesum-
mation of X
0
and X- densities, i.e.,
n
ð
x
Þ
=
n
ex
ð
x
Þ
+
n
tr
ð
x
Þ
. To include this,
we use the mass action model, which is expressed as follows:
n
tr
ð
x
Þ
=
n
ð
x
Þ
+
n
e
ð
x
Þ
+
n
A
ð
x
Þ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð
n
ð
x
Þ
+
n
e
ð
x
Þ
+
n
A
ð
x
ÞÞ
2
4
n
ð
x
Þ
n
e
ð
x
Þ
q
2
,
ð
3
Þ
where
n
A
(
x
)=
4
m
ex
m
e
π
_
m
tr
k
B
T
e
E
T
=
k
B
T
(
m
ex
,
m
tr
,and
m
e
are the masses of
X
0
, X-, and electron) is the relation of connecting concentrations of X
0
,
X-, and electrons by the law of mass action
6
,
33
,
34
.Figure
5
gshowsX
0
and
X- densities at the nanogap center as functions of
α
. As expected, the X-
density is zero for
α
= 0, i.e., there are no electrons; however, it con-
tinuously increases as
α
increases. Conversely, a decrease in the X
0
density effectively proves the occurrence of the exciton-to-trion
conversion process. Finally, we compare the experimental results from
the polarization control (Fig.
3
) and SLM control (Fig.
4
)tothe
theoretically analyzed data. Speci
fi
cally, we derive the X-/X
0
ratio from
04080120
2
3
1.8
2.0
2.2
1
2
3
4
PM on
PM off
SPP intensity (a. u.)
Time (s)
Energy (eV)
SPP intensity (a. u.)
1.80
1.85
1.90
0102030
1
2
Time (s)
X-/X
0
ratio
Energy (eV)
Norm. PL intensity (a. u.)
0
1
1.6
1.8
2.0
1
2
3
Raw
Fit
X
0
X-
1.6
1.8
2.0
1
2
3
4
PL intensity (a. u.)
PL intensity (a. u.)
Energy (eV)
Energy (eV)
PM off
PM on
a
bc
e
f
d
Adaptive excitation
MoS
2
e
e
e
e
e
e
e
e
e
e
e
h
e
h
e
h
e
h
e
e
e
h
e
e
h
e
e
h
e
On Off
Optimal PM
pump
~210 %
Spatial modulation of
SPP mode
Off
On
X
0
X
y
z
x
pump
Fig. 4 | All-optical control of exciton
–
trion interconversion. a
Illustration of
spatio-spectral modulation of SPP mode and excitonic emission response through
adaptive wavefront shaping.
b
Evolution of SPP intensity during stepwise sequence
feedback. Inset: optimized phase mask (PM) after sequence feedback.
c
SPP spectra
before (black) and after (red) wavefront shaping.
d
Corresponding PL spectra
before (left) and after (right) wavefront shaping
fi
tted to Voigt function. Black, blue,
and red lines indicate
fi
tofrawspectrum,X
0
, and X-, respectively. Black dots
indicate raw data.
e
Time-series normalized PL spectra during on/off switching of
optimal phase mask obtained in (
b
).
f
Corresponding X-/X
0
ratio.
Article
https://doi.org/10.1038/s41467-023-37481-1
Nature Communications
| (2023) 14:1891
5
the experimental results an
d match the corresponding
α
value to
estimate the experimental electron density localized at the nanogap
center in the SPP mode. As shown in Fig.
5
h, the X-/X
0
ratio from the
polarization control exhibits a positive correlation with the electron
density. By assuming ~100% exciton-to-trion conversion ef
fi
ciency
under the strain gradient
6
, we quantitatively estimate the electron
density based on the experimentally obtained X-/X
0
ratio, exhibiting a
maximum ten-fold enhancement of the localized electron density
35
.
Similarly, the SLM-controlled X-/X
0
ratio exhibits two distinct regions
assigned to the X-/X
0
ratio with and without the optimal phase mask.
Therefore, we con
fi
rm the local enhancement and con
fi
nement of
plasmon-induced hot electrons at the nanogap center in the SPP mode
and its effect on the additional exciton-to-trion conversion process.
Discussion
We developed an adaptive waveguide platform that enables the gen-
eration of high-purity trions, dynamic exciton
–
trion interconversion,
and their spatial modulation in 2D semiconductors. Speci
fi
cally, we
showed the precise controllability of the exciton-to-trion conversion
rate, which enables a dynamic transition between the dominant X
0
state and high-purity X- state via the modulation of the excitation
polarization. Furthermore, the spatial controllability of the SPP mode
was facilitated with adaptive wavefront shaping by the SLM, leading to
deterministic positioning of the exciton-to-trion conversion region.
Exploiting the drift-dominant exciton
fl
ux and converting con
fi
ned
excitons to trions through the nanoscale strain gradient result in the
ef
fi
cient harvesting of excitonic quasiparticles in the form of trions.
Unlike highly radiative trions in plasmonic cavity platforms, our high-
purity trions exhibit their intrinsic temporal characteristics, leading to
the high-ef
fi
ciency photoconversion
36
,
37
. Meanwhile, generating trionic
fl
ux with the converted trion should be a pressing matter, as it opens a
pathway toward manipulating excitonic/trionic
fl
ux ef
fi
ciently at the
nanoscale combined with existing plamonic MIM waveguide
geometry
38
–
40
.
Methods
Fabrication of nanogaps through focused ion-beam milling
Silicon wafers with thermally grown SiO
2
and a thickness of 1
μ
mwere
purchased from University Wafers, Boston, USA. Electron-beam eva-
poration was used to deposit 150-nm-thick Au on the wafers. An FEI
Nova 600 dual-beam system was used to perform focused ion beam
(FIB) milling on the wafers to etch into the Au and silica layers, creating
a nanogap. This part of the procedure was performed at an ion beam
voltage of 30 kV and current of 10 pA. This Au layer was then removed
using gold etchant (TFA, Transene Company Inc.), and a fresh layer of
Au (50 nm) was deposited onto the wafer using e-beam evaporation.
FIB milling was performed again at voltage 30 kV and a lower current (1
pA) to selectively etch the Au from the bottom of the nanogaps
41
.
Growth and transfer of MoS
2
MLs
A two-zone furnace was used to grow the MoS
2
ML
fl
akes; sulfur
fl
akes
(Merck,
≥
99.99%) were placed in the upstream zone; a 0.01 M sodium
molybdate aqueous solution was spun onto a SiO
2
/Si substrate as the
molybdenum precursor; the substrate was loaded into the down-
stream zone; the sulfur
fl
akes and substrate were heated at 200
∘
Cand
750
∘
C temperatures, respectively, for 7 min and maintained for 8 min;
the substrate was then cooled naturally to room temperature. The
entire process was performed with a N
2
carrier gas at a
fl
ow rate of 600
SCCM. The as-grown MoS
2
was then coated with poly (methyl
methacrylate) (PMMA) at 2500 rpm for 1 min. To delaminate the SiO
2
/
0.2
0.4
0.6
0
2
4
6
0
2
4
6
8
Fit
Pol. control
PM control
0
200
400
0
2
1.1
1.2
1.3
Exciton density
Electron density
0.0
0.5
1.0
0
1
2
X
0
X-
-200
0
200
0
50
100
150
MoS
2
ML
Nano-gap
Fit
0200400
0.5
1.0
x
(nm)
x
(nm)
x
(nm)
h
(nm)
n
gap
(
x
)
n
0
(
x
) (cm
-2
)
n
e
(
x
) (cm
-
2
)
10
14
10
16
Density (a. u.)
(cm
-2
)
(cm
-2
)
(cm
-2
)
X-/X
0
ratio (a. u.)
n
e
(
x
) (cm
-
2
)
-200
0
200
5.2
5.3
5.4
(eV)
x
(nm)
(eV)
h
(nm)
5.2
5.4
0
50
0
1.2
10
16
= 90º
= 45º
= 0º
PM off
PM on
Funneling
e
e
h
d
a
b
c
e
f
g
h
Nano-gap
200 nm
10
15
n
e
(0) (cm
-2
)
Fig. 5 | Theoretical investigation of electron funneling and exciton-to-trion
conversion dynamics.
Topography (
a
) and work function (
φ
)images(
b
)of
waveguide obtained by KPFM.
c
Work function pro
fi
le derived from dashed green
line in (
b
).
d
Height pro
fi
le of MoS
2
ML on nanogap of waveguide (green dots),
fi
tted line shape function (black line), and height pro
fi
le of nanogap without MoS
2
ML (yellow line).
e
Spatial density distribution of photoexcited excitons (black) and
electrons (red) under strain pro
fi
le estimated from
fi
tted line-shape function in (
d
).
f
Spatial electron density distribution as function of global defect density
α
for
estimated strain pro
fi
le.
g
X
0
density (blue) and X- (red) density as functions of
α
.
h
X-/X
0
ratio as function of
α
. Dashed black line represents theoretically obtained
fi
t
from (
g
). Dashed red line indicates theoretically driven electron density at center of
nanogap, i.e.,
n
e
(0) as function of
α
. Orange triangles and navy squares indicate
experimentally obtained values from Figs.
3
and
4
, respectively. Inset: closed view
of blue-
fi
lled region; values at bottom left are without optimal phase mask, whereas
those at top right are with optimal phase mask.
Article
https://doi.org/10.1038/s41467-023-37481-1
Nature Communications
| (2023) 14:1891
6