Supplementary Information for “Coherent optical wavelength
conversion via cavity-optomechanics”
Jeff T. Hill,
1
Amir H. Safavi-Naeini,
1
Jasper Chan,
1
and O. Painter
1
1
Kavli Nanoscience Institute and Thomas J. Watson,
Sr., Laboratory of Applied Physics,
California Institute of Technology, Pasadena, CA 91125
(Dated: October 9, 2012)
1
SUPPLEMENTARY FIGURES
2
λ
~
1,545
nm
a-m
VOA2
rf-sg
λ
-meter
λ
~
1,460
nm
λ
-mux
aom
a-m
VOA1
OMC
Tap
er
SW1
optical pr
e-amp
EDFA
power
meter
PD2
PD3
EIT reection
spectroscopy
LHe cryostat
mechanical mode / signal tone
spectroscopy and thermometry
input signal calibration
PD1
SW2
SW3
FPC
FPC
FPC
FPC
output laser (2)
input laser (1)
Supplementary Figure S1:
Detailed schematic of the experimental setup.
Two tunable,
external cavity diode lasers (ECDL) are used as control laser beams used to drive the wavelength
conversion process. The wavelength of both lasers is monitored and locked to an absolute frequency
of better than
±
5 MHz using a wavemeter (
λ
-meter). Both control beams can be amplitude mod-
ulated (a-m) using an RF signal generator (rf-sg), which produces the necessary optical sidebands
to perform EIT-like spectroscopy (see main text) and to generate the input signal for the wave-
length conversion process. An acousto-optic modulator (aom) is used to calibrate the input signal
(switching in the calibration signal with SW1). The light from both lasers are attenuated (VOA1,
VOA2) to the desired power, combined using a wavelength multiplexer (
λ
-mux), and then sent into
the dimpled optical fibre taper which is used to optically couple to the optomechanical crystal. The
optomechanical crystal sample and fibre taper are both located in a continuous-flow liquid helium
cryostat. Before entering the cryostat, the light passes through fibre polarization controllers (FPC)
to align the polarization of the light with that of the predominantly in-plane linear polarization of
the optomechanical crystal cavity modes. The light also passes through an optical switch (SW2),
allowing the light to be switched between either direction of propagation through the fibre taper,
and providing an accurate calibration of the losses in the taper before and after the optomechanical
crystal cavity (measuring the input-power-dependent optomechanical damping in both transmis-
sion directions allows one to separate the total loss of the taper into losses before and after the
cavity). The transmitted light from the cavity is either optically pre-amplified using an erbium
doped fibre amplifier (EDFA) and sent to a high-speed photodetector (PD2) for measurement of
the microwave power spectrum, or using optical switch (SW3), sent to an optical power meter
(PD3, power meter) for power calibration.
3
PSD (dBm/Hz)
Frequency (GHz)
ω
AOM
ω
1
-
ω
AOM
ω
1
+
ω
AOM
ω
1
4|A
1
A
3
|
2
(
β
1
/4)
2
|A
1
A
3
|
2
|A
1
|
4
β
1
2
cos(
φ
/4)
(
β
1
/4)
2
|A
1
A
3
|
2
Supplementary Figure S2:
Calibration tones.
Schematic of the four tones appearing in the elec-
tronic power spectrum of the photocurrent generated by the optical input signal and the additional
calibration signal. The three additional tones proportional to
A
3
are generated to calibrate the
input signal modulation index,
β
1
.
4
0.015
0.010
0
0.050
|S
21
|
2
2.0
1.5
1.0
0.5
0
0.020
C
2
/
C
1
b
a
1,000
3,000
0
2,000
n
c,2
60
62
64
66
68
T
0,1
(%)
0.090
0.095
0.105
0.100
0.110
η
1
Supplementary Figure S3:
Effect of large intracavity photon population.
(
a
), Recorded
coupling depth of the first order optical cavity as a function of the intracavity photon population
of the second order mode (green circles). The black curve represents a fit to the extracted coupling
depth. (
b
), Conversion efficiency as a function of
C
2
/C
1
. The dashed black line represents the
theoretical values from the nominal system parameters while the solid red line takes into account
the effect of the large intracavity photon population of cavity mode 2 (
n
c
,
2
). The blue circles
are the extracted data points and show much better correspondence to theory when this effect is
accounted for.
5
SUPPLEMENTARY METHODS
Theory of two-mode optomechanical system
We begin by considering the Hamiltonian describing the optomechanical interaction be-
tween two distinct optical modes (indexed
k
= 1
,
2) coupled to a shared mechanical mode
with annihilation operator
ˆ
b
,
ˆ
H
=
∑
k
̄
hδ
k
ˆ
a
†
k
ˆ
a
k
+ ̄
hω
m
ˆ
b
†
ˆ
b
+ (
ˆ
b
+
ˆ
b
†
)
∑
k
̄
hg
k
ˆ
a
†
k
ˆ
a
k
.
(S1)
Each optical mode has a frequency
ω
k
, and is driven by a laser at frequency
ω
l
,k
. The
Hamiltonian above is written in the interaction picture with
δ
k
=
ω
k
−
ω
l
,k
.
As shown below, strong driving at a mechanical frequency red detuned from each cav-
ity mode,
δ
k
∼
=
ω
m
, causes an effective beam splitter interaction to take place between the
mechanical mode and each optical cavity mode at an enhanced coupling rate
G
k
=
g
k
|
α
k
|
,
where
α
k
is the square root of the photon occupation in optical mode
k
. As long as this
coupling is weak with respect to the optical linewidths
κ
k
(
G
k
κ
k
), an adiabatic elimina-
tion of the optical cavities results in new effective mechanical loss rates
γ
OM
,k
into each of
the
k
optical degrees of freedom in the system. This “loss” can provide an effective coupling
between the optical cavity modes by allowing the exchange of excitations between the optical
resonances through the mechanical motion of the system. We calculate exactly what these
conversion rates are by using a scattering matrix formulation to understand the behaviour
of the system. Though completely general expressions can be derived, to best understand
the processes involved we focus on the parameter regime relevant to this experiment, i.e.
the weak-coupling, sideband-resolved case where
ω
m
κ
γ
OM
, and follow a scattering
matrix derivation of the induced optomechanical coupling between the optical waveguides
coupled to each cavity mode.
The linearized Heisenberg-Langevin equations of motion are found using Eq. (S1), the
displacement ˆ
a
k
→
α
k
+ ˆ
a
k
, and the inclusion of the optical input signal (ˆ
a
in
,k
(
t
)), optical
6
vacuum noise (ˆ
a
i
,k
(
t
)), and mechanical thermal fluctuation fields (
ˆ
b
in
(
t
)):
̇
ˆ
b
(
t
) =
−
(
iω
m
+
γ
i
2
)
ˆ
b
−
i
∑
k
G
k
(ˆ
a
+ ˆ
a
†
)
−
√
γ
i
ˆ
b
in
(
t
)
̇
ˆ
a
k
(
t
) =
−
(
iδ
k
+
κ
k
2
)
ˆ
a
k
−
iG
k
(
ˆ
b
+
ˆ
b
†
)
−
√
κ
e
,k
/
2ˆ
a
in
,k
(
t
)
−
√
κ
′
k
ˆ
a
i
,k
(
t
)
.
The mechanical loss rate
γ
i
=
ω
m
/Q
m
, determines the coupling of the system to the
thermal bath, and provides the most significant contribution in terms of noise processes
relevant to the performance of the optomechanical wavelength converter. The other loss
rates,
κ
k
,
κ
e,k
, and
κ
′
k
are, respectively for optical cavity mode
k
, the total optical loss rate,
the optical loss rate associated with coupling into the waveguide
k
, and parasitic optical
loss rate into all other channels that are undetected representing a loss of information. Due
to the local evanescent side-coupling scheme studied here (which bi-directionally couples to
the cavity mode),
κ
k
=
κ
e
,k
/
2 +
κ
′
k
, i.e. only a fraction
η
k
≡
κ
e
,k
/
2
κ
k
of the photons leaving
the cavity get detected and
η
k
≤
1
/
2.
The equations of motion are linear and thus the system can be analyzed more simply in
the frequency domain. Solving for the spectrum of the mechanical mode
ˆ
b
(
ω
) in terms of
the input noise operators, we find:
ˆ
b
(
ω
) =
−
√
γ
i
ˆ
b
in
(
ω
)
i
(
ω
m
−
ω
)+
γ/
2
+
∑
k
iG
k
i
(
δ
k
−
ω
)+
κ
k
/
2
√
κ
e
,k
/
2
ˆ
a
in
,k
(
ω
)+
√
κ
′
k
ˆ
a
i
,k
(
ω
)
i
(
ω
m
−
ω
)+
γ/
2
+
∑
k
iG
k
−
i
(
δ
k
+
ω
)+
κ
k
/
2
√
κ
e
,k
/
2
ˆ
a
†
in
,k
(
ω
)+
√
κ
′
k
ˆ
a
†
i
,k
(
ω
)
i
(
ω
m
−
ω
)+
γ/
2
,
where the mechanical frequency
ω
m
is now modified by the optical spring, and the mechanical
linewidth is given by
γ
≡
γ
i
+
γ
OM
,
1
+
γ
OM
,
2
. The optomechanical damping terms
γ
OM
,k
come
from coupling of the mechanical system to the optical mode
k
, and are given by the relation
γ
OM
,k
= 2
|
G
k
|
2
Re
[
1
i
(
δ
k
−
ω
m
) +
κ
k
/
2
−
1
−
i
(
δ
k
+
ω
m
) +
κ
k
/
2
]
=
4
|
G
k
|
2
κ
k
for
δ
k
≈
ω
m
.
(S2)
7
The last expression is a simplification that is often made, and is equivalent to looking at
spectral properties at detunings from the mechanical frequency much smaller than
κ
(so the
optical lineshape isn’t taken into account). Under this approximation, the optical spectrum
is
κ
k
2
ˆ
a
k
(
ω
) =
iG
k
√
γ
i
ˆ
b
in
(
ω
)
i
(
ω
m
−
ω
)+
γ/
2
+
∑
j
2
G
j
G
k
κ
j
√
κ
e
,j
/
2
ˆ
a
in
,j
(
ω
)+
√
κ
′
j
ˆ
a
i
,j
(
ω
)
i
(
ω
m
−
ω
)+
γ/
2
+
∑
j
iG
j
G
k
2
ω
m
√
κ
e
,j
/
2
ˆ
a
†
in
,j
(
ω
)+
√
κ
′
j
ˆ
a
†
i
,j
(
ω
)
i
(
ω
m
−
ω
)+
γ/
2
−
√
κ
e
,k
/
2ˆ
a
in
,k
(
ω
)
−
√
κ
′
k
ˆ
a
i
,k
(
ω
)
.
(S3)
From this expression, we see that there are several noise operators incident on optical mode
k
. The thermal fluctuations from the environment,
ˆ
b
in
(
ω
), are converted into noise photons
over the mechanical bandwidth
γ
. There is also an induced coupling to the optical mode
j
6
=
k
, and photons originally incident only on mode
j
, i.e. ˆ
a
in
,j
(
ω
) are now also coupled into
ˆ
a
k
. Finally we note that vacuum noise creation operators are also present in this expression.
These give rise to quantum noise through spontaneous emission of phonons, though we show
below that as
ω
m
is made very large with respect to
κ
k
, these terms diminish in importance.
From outside the optomechanical system, we only have access to photons sent into
the system, ˆ
a
in
,j
, and those leaving the system, ˆ
a
out
,k
, on transmission. The relation be-
tween these operators is best understood through scattering parameters, and can be de-
rived using the equation for ˆ
a
k
(
ω
) (Eq. S3) and the input-output boundary conditions
ˆ
a
out
,k
(
ω
) = ˆ
a
in
,k
(
ω
) +
√
κ
e
,k
/
2ˆ
a
k
(
ω
). After some algebra, the output operator is expressed
as
ˆ
a
out
,k
(
ω
) =
s
th
,k
(
ω
)
ˆ
b
in
(
ω
)
+
t
k
(
ω
)ˆ
a
in
,k
(
ω
) +
s
kj
(
ω
)ˆ
a
in
,j
(
ω
)
+
∑
m
n
opt
,m
(
ω
)ˆ
a
i
,m
(
ω
)
+
∑
m
s
adj,in
,m
(
ω
)ˆ
a
†
in
,m
(
ω
)
+
∑
m
s
adj,i
,m
(
ω
)ˆ
a
†
i
,m
(
ω
)
(S4)
The scattering coefficient
s
th
,k
(
ω
) is the conversion efficiency of mechanical thermal noise to
8
photons, and is given by
s
th
,k
(
ω
) =
i
√
η
k
√
γ
i
γ
OM
,k
i
(
ω
m
−
ω
) +
γ/
2
.
(S5)
The coefficient
t
k
(
ω
) in our system is transmission amplitude, given by
t
k
(
ω
) = (1
−
2
η
k
) +
η
k
γ
OM
,k
i
(
ω
m
−
ω
) +
γ/
2
.
(S6)
In principle, this coefficient is the electromagnetically induced transparency transmission
coefficient, though here it is written about the mechanical frequency for detunings on the
order of
γ
and the
κ
’s are too large (
κ
γ
) for the optical lineshape to be considered in
the expression. This is the expected result of the weak coupling approximation. Finally, the
most important coefficient is the wavelength conversion coefficient
s
kj
(
ω
) which is given by
s
kj
(
ω
) =
√
η
k
η
j
√
γ
OM
,k
γ
OM
,j
i
(
ω
m
−
ω
) +
γ/
2
.
(S7)
Efficiency, Bandwidth, and Noise
We calculate the spectral density of the output field on port
k
(
S
out
,k
(
ω
)) assuming an
input field spectral density on the opposing optical port
j
(
S
in
,j
(
ω
)), vacuum inputs on all
other optical channels, and thermal noise from a phonon bath with thermal occupation
n
b
.
The spectral densities here have units of photons
/
Hz
·
s and can be interpreted as photon flux
per unit bandwidth. At first we ignore the field creation operators in the scattering relation
Eq. (S4) (which give rise to quantum noise and can be neglected when the mechanical
resonator occupation number is greater than one) and arrive at the following expression:
S
out
,k
(
ω
) =
∫
∞
−∞
d
ω
′
〈
ˆ
a
†
out
,k
(
ω
)ˆ
a
out
,k
(
ω
′
)
〉
=
∫
∞
−∞
s
∗
th
,k
(
−
ω
)
s
th
,k
(
ω
′
)
〈
ˆ
b
†
in
(
ω
)
ˆ
b
in
(
ω
′
)
〉
+
t
∗
k
(
−
ω
)
t
k
(
ω
′
)
〈
ˆ
a
†
in
,k
(
ω
)ˆ
a
in
,k
(
ω
′
)
〉
+
s
∗
kj
(
−
ω
)
s
kj
(
ω
′
)
〈
ˆ
a
†
in
,j
(
ω
)ˆ
a
in
,j
(
ω
′
)
〉
+
n
∗
opt
,k
(
−
ω
)
n
opt
,k
(
ω
′
)
〈
ˆ
a
†
i
,k
(
ω
)ˆ
a
i
,k
(
ω
′
)
〉
+
n
∗
opt
,j
(
−
ω
)
n
opt
,j
(
ω
′
)
〈
ˆ
a
†
i
,j
(
ω
)ˆ
a
i
,j
(
ω
′
)
〉
d
ω
′
.
9
This reduces to,
S
out
,k
(
ω
) =
η
k
γ
i
γ
OM
,k
(
ω
+
ω
m
)
2
+ (
γ/
2)
n
b
+
η
k
η
j
γ
OM
,k
γ
OM
,j
(
ω
+
ω
m
)
2
+ (
γ/
2)
2
S
in
,j
(
ω
)
,
accounting for the noise autocorrelation functions. From here, we see that the maximum
photon wavelength conversion efficiency is given by
η
max
=
η
1
η
2
4
γ
OM
,
1
γ
OM
,
2
(
γ
i
+
γ
OM
,
1
+
γ
OM
,
2
)
2
.
(S8)
The conversion process has a bandwidth (BW) equal to the mechanical linewidth
BW
≡
γ
i
+
γ
OM
,
1
+
γ
OM
,
2
.
(S9)
From the same expression, we calculate a value for the optical signal to noise ratio (OSNR),
to find
OSNR
classical
kj
=
η
j
γ
OM
,j
γ
i
n
b
S
in
,j
(
ω
)
.
(S10)
Equation (S10) only takes into account the thermal noise in the system and therefore
holds only when
γ
i
n
b
/γ
OM
,j
1. At higher powers, the quantum noise processes which give
rise to the back-action limit in cooling also give rise to excess noise for wavelength conversion.
To consider these terms, the creation operators in Eq. (S4) must not be neglected and the
full relation is then found to be
S
out
,k
(
ω
) =
η
k
γ
OM
,k
(
ω
+
ω
m
)
2
+ (
γ/
2)
2
γ
i
n
b
+
η
k
γ
OM
,k
(
ω
+
ω
m
)
2
+ (
γ/
2)
2
γ
OM
,k
(
κ
k
4
ω
m
)
2
+
η
k
γ
OM
,k
(
ω
+
ω
m
)
2
+ (
γ/
2)
2
γ
OM
,j
(
κ
j
4
ω
m
)
2
+
η
k
η
j
γ
OM
,k
γ
OM
,j
(
ω
+
ω
m
)
2
+ (
γ/
2)
2
S
in
,j
(
ω
)
.
(S11)
The quantum limited OSNR is then found to be
OSNR
kj
=
η
j
γ
OM
,j
S
in
,j
(
ω
)
γ
i
n
b
+
γ
OM
,k
(
κ
k
4
ω
m
)
2
+
γ
OM
,j
(
κ
j
4
ω
m
)
2
.
(S12)
For the quantum back-action terms to become significant as compared to the thermal noise
on the mechanical system, one needs
γ
i
n
b
/γ
∼
(
κ
4
ω
m
)
2
. This is a regime that is yet to be
10
reached in experiments and as such the quantum back-action noise terms can be neglected
for experiments to date. Perhaps more importantly, we ask “what is the amount of noise
added to a signal of a single photon?”. We can express the signal to noise ratios as OSNR
kj
=
S
in
,j
(
ω
)
/n
added
, with
n
added
=
γ
i
n
b
+
γ
OM
,k
(
κ
k
4
ω
m
)
2
+
γ
OM
,j
(
κ
j
4
ω
m
)
2
η
j
γ
OM
,j
.
(S13)
Assuming
κ
j
=
κ
k
=
κ
,
γ
i
γ
, and
γ
OM
,j
=
γ
OM
,k
=
γ
OM
, this expression becomes
n
added
≈
2
η
−
1
j
(
γ
i
n
b
γ
+
(
κ
4
ω
m
)
2
)
.
(S14)
To achieve the threshold where a single photon incoming on port
j
is converted, and is
more likely to be detected on port
k
than an upconverted thermal excitation, we require
n
added
<
1. This requirement is equivalent to stating that in the absence of signals on port
j
and
k
, a cooled phonon occupation of ̄
n
=
η
j
/
2 must be achieved. We note that
η
k
does not
make an appearance in this equation, since reduced output coupling efficiency attenuates
the thermal noise and signal equivalently. For the same reason, 1
/η
j
is present in the noise
quanta expression. The factor of two can be understood to come from the fact that for
good conversion, we require the photon to enter, and exit the system before a phonon is
up converted. The former two processes happen at
γ
OM
, while the latter is given by the
thermalization rate
γ
i
n
b
.
11