of 7
Towards Visible Soliton Microcomb Generation
Seung Hoon Lee
1
*, Dong Yoon Oh
1
*, Qi-Fan Yang
1
*, Boqiang Shen
1
*, Heming Wang
1
*,
Ki Youl Yang
1
, Yu Hung Lai
1
, Xu Yi
1
, and Kerry Vahala
1
1
T. J. Watson Laboratory of Applied Physics, California Institute of Technology, Pasadena, California 91125, USA.
*These authors contributed equally to this work.
Corresponding author: vahala@caltech.edu
Frequency combs have applications that ex-
tend from the ultra-violet into the mid infrared
bands. Microcombs
1
, a miniature and often
semiconductor-chip-based device, can potentially
access most of these applications, but are cur-
rently more limited in spectral reach. Here,
we demonstrate mode-locked silica microcombs
with emission near the edge of the visible spec-
trum. By using both geometrical and mode-
hybridization dispersion control, devices are en-
gineered for soliton generation while also main-
taining optical Q factors as high as 80 million.
Electronics-bandwidth-compatible (20 GHz) soli-
ton mode locking is achieved with threshold pow-
ers as low as 5.4 mW. These are the shortest
wavelength soliton microcombs demonstrated to
date and could be used in miniature optical-
clocks
2,3
. The results should also extend to visible
and potentially ultraviolet bands.
Soliton microcombs
4–8
provide a pathway to miniatur-
ize many conventional comb applications. They have
also opened investigations into new nonlinear physics
associated with dissipative Kerr solitons
4
and Stokes
solitons
9
. In contrast to early microcombs, soliton micro-
combs eliminate instabilities, provide stable (low-phase-
noise) mode locking and feature a highly reproducible
spectral envelope. Many applications of these devices
are being studied including chip-based optical frequency
synthesis
10
, secondary time standards
3
and dual-comb
spectroscopy
11–13
.
Also, a range of operating wave-
lengths is opening up by use of several low-optical-loss
dielectric materials for resonator fabrication. In the near-
infrared (IR), microcombs based on magnesium fluoride
4
,
silica
5,14
and silicon nitride
6–8,15,16
are being studied for
frequency metrology and frequency synthesis. In the
mid-IR spectral region silicon nitride
17
, crystalline
18
, and
silicon-based
19
Kerr microcombs as well as quantum-
cascade microcombs
20
are being studied for application
to molecular fingerprinting.
At shorter wavelengths below 1
μ
m, microcomb tech-
nology would benefit optical atomic clock technology
21
,
particularly efforts to miniaturize these clocks. For ex-
ample, microcomb optical clocks based on the D1 transi-
tion (795 nm) and the two-photon clock transition
22
(798
nm) in rubidium have been proposed
2,3
. Also, a micro-
comb clock using two-point locking to rubidium D1 and
D2 lines has been demonstrated
23
by frequency doubling
from the near-IR. More generally, microcomb sources in
the visible and ultraviolet bands could provide a minia-
ture alternative to larger mode-locked systems such as
titanium sapphire lasers in cases where high power is not
required. It is also possible that these shorter wavelength
systems could be applied in optical coherence tomogra-
phy systems
24–26
. Efforts directed towards short wave-
length microcomb operation include 1
μ
m microcombs in
silicon nitride microresonators
27
as well as harmonically-
generated combs. The latter have successfully converted
near IR comb light to shorter wavelength bands
28
and
even into the visible band
29,30
within the same res-
onator used to create the initial comb of near-IR frequen-
cies. Also, crystalline resonators
31
and silica microbubble
resonators
32
have been dispersion-engineered for comb
generation in the 700 nm band. Finally, diamond-based
microcombs afford the possibility of broad wavelength
coverage
33
. However, none of the short wavelength mi-
crocomb systems have so far been able to generate stable
mode-locked microcombs as required in all comb appli-
cations.
A key impediment to mode-locked microcomb opera-
tion at short wavelengths is material dispersion associ-
ated with the various dielectric materials used for mi-
croresonator fabrication. At shorter wavelengths, these
materials feature large normal dispersion that dramat-
ically increases into the visible and ultraviolet bands.
Soliton-based mode-locking, on the other hand, requires
anomalous dispersion. Dispersion engineering by proper
design of the resonator geometry
2,31,32,34–39
offers a pos-
sible way to offset the normal dispersion. Typically, by
compressing a resonator’s waveguide dimension, geomet-
ric dispersion will ultimately compensate a large normal
material dispersion component to produce overall anoma-
lous dispersion. For example, in silica, strong confine-
ment in bubble resonators
32
and straight waveguides
40
has been used to push the anomalous dispersion transi-
tion wavelength from the near-IR into the visible band.
Phase matching to ultraviolet dispersive waves has also
been demonstrated using this technique
40
.
However,
to compensate the rising material dispersion this com-
pression must increase as operational wavelength is de-
creased, and as a side effect highly-confined waveguides
tend to suffer increased optical losses. This happens
because mode overlap with the dielectric waveguide in-
terface is greater with reduced waveguide cross section.
Consequently, the residual fabrication-induced roughness
of that interface degrades the resonator Q factor and in-
creases pumping power (i.e., comb threshold power varies
arXiv:1705.06703v2 [physics.optics] 29 May 2017
2
t (
μ
m)
700
1000
1300
1600
2
4
6
8
(TM1)
Anomalous
Normal
II
I
III
λ
λ
(nm)
ZDW
Bulk silica
pump at 1550 nm
pump at 1064 nm
pump at 778 nm
I
b
e
d
a
II
III
c
t
185
190
195
Power (20 dB/div)
260
280
380
385
390
Frequency (THz)
0
10
20
P
th
(mW)
0
100
200
300
Q (M)
2
4
6
8
0
20
40
t (
μ
m)
A
eff
(
μ
m
2
)
778
1064
1550
λ
(nm)
pump
FIG. 1:
Soliton frequency comb generation in dispersion-engineered silica resonators.
(
a
) A render-
ing of a silica resonator with the calculated mode profile of the TM1 mode superimposed. (
b
) Regions of normal
and anomalous dispersion are shown versus silica resonator thickness (t) and pump wavelength. The zero disper-
sion wavelength (
λ
ZDW
) for the TM1 mode appears as a blue curve. Plot is made for a 3.2-mm-diameter silica res-
onator. Three different device types I, II, and III (corresponding to three oxide thicknesses) are indicated for soliton
generation at 1550 nm, 1064 nm and 778 nm. (
c
) Measured Q factors and comb threshold powers versus thickness
and pump wavelength for the three device types. Effective mode area (A
eff
) of the TM1 mode family is also plotted
as a function of wavelength and thickness. (
d
) A photograph of a silica resonator (Type III device pumped at 778
nm) while generating a soliton stream. The pump light is coupled via a tapered fiber from the left side of the res-
onator. The red light along the circumference of the resonator and at the right side of the taper is due to the short
wavelength components of the soliton comb. (
e
) Soliton frequency comb spectra measured using device types I, II,
and III designed for pump wavelengths 1550 nm, 1064 nm, and 778 nm, respectively. Pump frequency location is
indicated by a dashed vertical line. The soliton pulse repetition rate of all devices is about 20 GHz. Insets: cross-
sectional SEM images of the fabricated resonators. White scale bar is 5
μ
m.
inverse quadratically with Q factor
41
).
Minimizing material dispersion provides one way to
ease the impact of these constraints. In this sense, sil-
ica offers an excellent material for short wavelength op-
eration, because it has the lowest dispersion among all
on-chip integrable materials. For example, at 778 nm,
silica has a group velocity dispersion (GVD) equal to 38
ps
2
/km, which is over 5X smaller than the GVD of silicon
nitride at this wavelength (
>
200 ps
2
/km)
42
. Other inte-
grable materials that are also transparent in the visible,
such as diamond
33
and aluminum nitride
43
, have disper-
sion that is similar to or higher than silicon nitride. Sil-
ica also features a spectrally-broad low-optical-loss win-
dow so that optical Q factors can be high at short wave-
lengths. Here we demonstrate soliton microcombs with
pump wavelengths of 1064 nm and 778 nm. These are the
shortest soliton microcomb wavelengths demonstrated to
date. By engineering geometric dispersion and by em-
ploying mode hybridization, net anomalous dispersion
has been achieved at these wavelengths while also main-
taining high optical Q factors (80 million at 778 nm, 90
million at 1064 nm). The devices have large (millimeter-
scale) diameters and produce single soliton pulse streams
at rates that are both detectable and processible by low-
cost electronic circuits. Besides illustrating the flexibil-
ity of silica for soliton microcomb generation across a
3
range of short wavelengths, these results are relevant to
potential secondary time standards based on transitions
in rubidium
2,3,23
. Using dispersive-wave engineering in
silica it might also be possible to extend the emission of
these combs into the ultraviolet as recently demonstrated
in compact silica waveguides
40
.
The silica resonator used in this work is shown
schematically in Fig. 1a. A fundamental mode profile
is overlaid in the cross-sectional rendering. As described
in detail below, the resonator thickness (t) is controlled
to obtain net anomalous dispersion at the design wave-
lengths. The resonator design is a variation on the wedge
resonator
44
and precise thickness control (t-control) is
possible because this layer is formed through oxidation
of a silicon wafer. The diameter of all resonators in this
work (and the assumed diameter in all simulations) is 3.2
mm, which corresponds to a free-spectral-range (FSR) of
approximately 20 GHz. Further details on fabrication are
given elsewhere
44
. As an aside, we note that a waveguide-
integrated version of this design is also possible
45
. Adap-
tation of that device using the methods described here
would enable full integration with other photonic ele-
ments on the silicon chip.
Fig. 1b illustrates the dispersion design space by show-
ing regions of anomalous and normal dispersion for the
TM1 mode family versus resonator thickness t and pump-
ing wavelength. The plot shows that as the pump wave-
length decreases the resonator needs to be thinner to ac-
cess the anomalous dispersion regime. With this in mind,
we have selected three different device types for soliton
frequency comb operation at three different pump wave-
lengths. These are indicated in Fig. 1b as I, II and III
with colored dots. At a pump wavelength of 1550 nm, the
anomalous dispersion window is wide because bulk sil-
ica possesses anomalous dispersion at wavelengths above
1270 nm. For this device (Type I), an 8-
μ
m thickness was
used. Devices of type II and III have thicknesses near 3.5
μ
m and 1.5
μ
m for operation with pump wavelengths of
1064 nm and 778 nm, respectively. Measured Q factors
for the three device types are plotted in the upper panel
of Fig. 1c. Maximum Q factors at thicknesses which also
produce anomalous dispersion were: 280 million (Type I,
1550 nm), 90 million (Type II, 1064 nm) and 80 million
(Type III, 778 nm).
Using these three designs, soliton frequency combs
were successfully generated with low threshold pump
power. Shown in Fig. 1d is a photograph of a type III de-
vice. Soliton frequency components in the 700 nm band
generate the red light in the photograph. Fig. 1e shows
optical spectra of the soliton microcombs generated for
each device type. A slight Raman-induced soliton self-
frequency-shift is observable in the type I and type II
devices
5,46–48
. Scanning electron microscope (SEM) im-
ages appear as insets in Fig.
1e and provide cross-
sectional views of the three device types. It is worthwhile
to note that microcomb threshold power, expressed as
P
th
A
eff
p
Q
2
(
λ
p
is pump wavelength and
A
eff
is ef-
fective mode area) remains within a close range of powers
for all devices (lower panel of Fig. 1c). This can be un-
derstood to result from partial compensation of reduced
Q factor in the shorter wavelength devices by reduced op-
tical mode area (see plot in Fig. 1c). For example, from
1550 nm to 778 nm the mode area is reduced by roughly
a factor of 9 and this helps to offset a decrease in Q fac-
tor of 3X. The resulting
P
th
increase (5.4 mW at 778 nm
versus approximately 2.5 mW at 1550 nm) is therefore
caused primarily by the decrease in pump wavelength
λ
p
.
In the following sections additional details on the device
design, dispersion and experimental techniques used to
generate these solitons are presented.
Dispersion simulations for TM modes near 1064 nm
are presented in Fig. 2a and show that TM modes with
anomalous dispersion occur in silica resonators having
oxide thicknesses less than 3.7
μ
m. Aside from the thick-
ness control, a secondary method to manipulate disper-
sion is by changing the wedge angle (see Fig. 2a). Here,
wedge angles between 30 and 40 degrees were chosen
in order to maximize the Q factors
44
. The resonator
dispersion is characterized by measuring mode frequen-
cies using a scanning external-cavity-diode-laser (ECDL)
whose frequency is calibrated using a Mach-Zehnder in-
terferometer. As described elsewhere
4,5
the mode fre-
quencies,
ω
μ
, are Taylor expanded to second order as
ω
μ
=
ω
o
+
μD
1
+
μ
2
D
2
/
2, where
ω
o
denotes the pumped
mode frequency and
D
1
/
2
π
is the FSR.
D
2
is related to
the GVD,
β
2
, by
D
2
=
cD
2
1
β
2
/n
o
where
c
and
n
o
are
the speed of light and material refractive index, respec-
tively. The measured frequency spectrum of the TM1
mode family in a 3.4
μ
m thick resonator is plotted in Fig.
2b. The plot gives the frequency as relative frequency
(i.e.,
ω
μ
ω
o
μD
1
) to make clear the second-order dis-
persion contribution. Also shown is a fitted parabola (red
curve) revealing
D
2
/
2
π
= 3
.
3 kHz (positive parabolic
curvature indicates anomalous dispersion). Some avoided
mode crossings are observed in the spectrum. The dis-
persion measured in resonators of different thicknesses,
marked as solid dots in Fig. 2a, is in good agreement
with numerical simulations.
The experimental setup for generation of 1064 nm
pumped solitons is shown in Fig. 2c. The microres-
onator is pumped by a CW laser amplified by a YDFA.
The pump light and comb power are coupled to and from
the resonator by a tapered fiber
49,50
. Solitons are gener-
ated while scanning the laser from higher frequencies to
lower frequencies across the pump mode
4–6
. The pump
light is modulated by an electro-optic PM to overcome
the thermal transient during soliton generation
5,6,51
. A
servo control referenced to the soliton power is employed
to capture and stabilize the solitons
51
. Shown in Fig.
2d are the optical spectra of solitons pumped at 1064
nm. These solitons are generated using the mode family
whose dispersion is characterized in Fig. 2b. Due to the
relatively low dispersion (small
D
2
), these solitons have a
short temporal pulsewidth. Using the hyperbolic-secant-
squared fitting method (see orange and green curves in
Fig. 2d) a soliton pulse width of 52 fs is estimated for