ARTICLE

https://doi.org/10.1038/s41467-022-31134-5 OPEN

Ultra-low-power second-order nonlinear optics on

a chip
Timothy P. McKenna1,3,5, Hubert S. Stokowski 1,5, Vahid Ansari 1, Jatadhari Mishra1, Marc Jankowski1,2,
Christopher J. Sarabalis1,4, Jason F. Herrmann1, Carsten Langrock1, Martin M. Fejer1 & Amir H. Safavi-Naeini 1 ✉
1234567890():,;

Second-order nonlinear optical processes convert light from one wavelength to another and
generate quantum entanglement. Creating chip-scale devices to efficiently control these
interactions greatly increases the reach of photonics. Existing silicon-based photonic circuits
utilize the third-order optical nonlinearity, but an analogous integrated platform for second-
order nonlinear optics remains an outstanding challenge. Here we demonstrate efficient
frequency doubling and parametric oscillation with a threshold of tens of micro-watts in an
integrated thin-film lithium niobate photonic circuit. We achieve degenerate and non-
degenerate operation of the parametric oscillator at room temperature and tune its emission
over one terahertz by varying the pump frequency by hundreds of megahertz. Finally, we
observe cascaded second-order processes that result in parametric oscillation. These reso-
nant second-order nonlinear circuits will form a crucial part of the emerging nonlinear and
quantum photonics platforms.

1 E.L. Ginzton Laboratory, Stanford University, Stanford, CA 94305, USA. 2 Physics & Informatics Laboratories, NTT Research, Inc., Sunnyvale, CA 94085,

USA. 3Present address: Physics & Informatics Laboratories, NTT Research, Inc., Sunnyvale, CA 94085, USA. 4Present address: Flux Photonics Inc., 580 Crespi
Dr Unit R, Pacifica, CA 94044, USA. 5These authors contributed equally: Timothy P. McKenna, Hubert S. Stokowski. ✉email: safavi@stanford.edu

NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5

T
he remarkable progress and impact of silicon photonics has In this work, we make waveguides from a thin film of X-cut
led to the development of complex and high-performance lithium niobate (Fig. 1b), which has its largest electro-optic and
optical systems for communications, sensing, and quantum χ(2) tensor components parallel to the surface of the chip. This
and classical information processing. In addition to linear pas- orientation has been used in recent demonstrations of tele-
sives, modulators, and detectors, many applications would sig- communications modulators25,27, frequency combs26, cryogenic
nificantly benefit from versatile nonlinearities. Integrated frequency converters32–34, and sources exhibiting quantum
photonic circuits made with centrosymmetric silicon1,2 or correlations35,36, which form an emerging thin-film LN platform.
amorphous silicon nitride3–5 confine light in dispersion- We use magnesium oxide (MgO) doped lithium niobate to sup-
engineered waveguides and resonators to enhance the third- press pump-induced absorption and reduce the photorefractive
order optical nonlinearity and have been used successfully to damage typically experienced by devices fabricated with undoped
demonstrate optical frequency combs6–8, wavelength conversion9, congruently grown lithium niobate37.
and squeezed light generation10,11. Efforts continue to further
improve the efficiency and tailorability of these devices by
incorporating second-order nonlinearity to enable stronger Results
interactions at lower power and reduce the number of competing Due to both its geometry and material properties, the dispersion
nonlinear processes that emerge. Second-order nonlinearity can of the waveguide introduces a phase velocity mismatch propor-
be introduced by breaking the symmetry of a crystal12,13 or tional to Δn—the difference in refractive indices between fun-
heterogeneously integrating a non-centrosymmetric material14,15. damental (FH) and second harmonic (SH) modes as shown in
Alternatively, photonic circuits may be built directly from a χ(2) Fig. 1c. To achieve efficient nonlinear interactions, we compen-
nonlinear material such as aluminum nitride16 or lithium niobate sate for the phase velocity mismatch by periodically poling the
(LN). LN can be periodically poled to compensate for phase mis- LN crystal. This quasi-phase-matching technique provides
match due to dispersion17–23 and supports high-Q optical momentum conservation and enables the use of the same fun-
resonances24, a large electro-optic coefficient25–27, and Kerr damental transverse electric (TE) spatial mode at both
nonlinearity28,29. Here we show ultra-efficient resonant χ(2) non- wavelengths19,20. These modes exhibit the tightest confinement
linear optical functions (Fig. 1a) on a chip that incorporates quasi- and have the strongest overlap with the large d33 component of
phase-matching with a nonlinear optical resonator. We overcome the χ(2) nonlinear tensor, thereby enabling a large nonlinear
parasitic effects that so far have limited the stability and perfor- interaction rate. We use a poling period of Λ = λSH/Δn ≈ 3.7 μm.
mance of integrated LN devices to demonstrate second-order pro- The inset of Fig. 1d shows a second-harmonic microscope pic-
cesses such as optical parametric oscillation, which have previously ture of the periodic poling before waveguide fabrication. We
only been observed in LN bulk resonators30,31. We operate an observe the formation of oblong shapes with greyscale fringes
optical parametric oscillator (OPO) across degenerate and non- between finger electrodes (black) that correspond to inverted
degenerate regimes and show tuning of the emission spectrum crystal domains38.
across one THz by adjusting the frequency of the pump across The waveguide forms a racetrack resonator with a straight
hundreds of MHz, all at room temperature. We also show fre- section length L of 3.2 mm (see Fig. 1d) that supports resonances
quency doubling that leads to highly-enhanced effective third-order across a broad range of wavelengths. We employ a phase-
nonlinearity, resulting in cascaded parametric oscillation. The pre- mismatched waveguide coupler design to efficiently couple light
sented coupled-mode theory accurately models the dynamics and into the resonator at both FH and SH frequencies39. Near the FH
confirms the operating modes of the device. and SH frequencies, we measure intrinsic quality factors

Fig. 1 Integrated, resonant second-order nonlinear optical device. a Schematic of a resonant second-order nonlinear optical device. Driving the cavity
with second harmonic light (blue) results in optical parametric oscillation at the fundamental; Driving at the fundamental frequency generates second
harmonic light. b Periodically-poled lithium niobate ridge waveguide that confines light to a small volume and supports nonlinear interactions. The
transverse electric field parallel to the surface of the chip is plotted for the fundamental spatial mode. c Effective index of the waveguide spatial modes as a
function of wavelength. Periodic poling compensates for the phase velocity mismatch (∝Δn) between fundamental and SH modes. d The racetrack
resonator used as a platform for nonlinear optics. Laser light is injected through an evanescent coupler on the top and undergoes nonlinear interaction in
the bottom, periodically-poled section. The resonator is not coupled to the waveguide below it, which serves as a tool for poling diagnostics. The laser
confocal microscope picture has been colorized; blue shading highlights the poling electrodes location during the fabrication process. Inset shows a second-
harmonic microscope picture of the poled region. Inverted domains stretch between black electrode fingers.

2 NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5 ARTICLE

a ℏΩk∣Bk∣2, and evolve in time as

laser
VOA 775 nm d κA;m
Am ¼  A  2i ∑ g k;nm An Bk eiδk;nm t ð1Þ
VOA EDFA
laser OSA dt 2 m kn
1550 nm
d κB;k
FPC FPC
InGaAs B ¼ B  i ∑ g k;nm Am An eþiδk;nm t ; ð2Þ
50% APD dt k 2 k mn
PM PPLN VOA with δk,nm ≡ Ωk − ωn − ωm. To operate as an optical parametric
5%
lensed oscillator (OPO), a laser driving term is added to the first equa-
fiber
lensed
VOA tion, while adding a laser driving term to the second equation
fibers
dichroic Si APD
causes second harmonic generation (SHG) and eventually
5%
LNOI chip mirror operation as a cascaded OPO.
b PM
Optical parametric oscillation occurs when the second-
0 0 harmonic mode is driven to a sufficiently large steady-state cav-
ity occupation ∣B0∣2. The system will begin to oscillate at this

Pump Trans. (dB)

SH power (dBm)

-10
-10 input power, either as a degenerate OPO with emission into ω0
mode or as a nondegenerate OPO emitting into a pair of modes
-20
-20
ω±m. The mode of oscillation is that with the lowest threshold
Pth,m, which strongly depends on laser detuning Δ, modal
-30
detuning μ, total loss κ, extrinsic loss κ(e), and dispersion ζ2m2:
-30 
2 
1542 1544 1546 1548 1550 1552 1554 _Ω0 1
c Pump wavelength (nm) Pth;m ¼ Δ 2
þ κ =2
16jg 0;mm j2 κðeÞ B;0
0


B;0
 ð3Þ

Pump Trans. (dB)
OPO power (dBm)

2 2
0
´ Δ þ μ  ζ 2 m þ κA;m κA;m :
-5
-10 The pair of modes ω±m with the lowest loss rates will experi-
-20
-10 ence the lowest threshold and oscillate first as we increase the
-30 pump power. Above the threshold, the OPO output power fol-
-40 lows a square-root function of the input power PB,0 provided that
765 765.5 766 766.5 767 767.5 the input power is not sufficiently large to produce simultaneous
Pump wavelength (nm) oscillation of multiple mode pairs:
sffiffiffiffiffiffiffiffiffiffi !
Fig. 2 Characterization of nonlinear devices. a Experimental setup. We 4ηB;0
 PB;0
couple light at around 775 or 1550 nm wavelengths onto the chip by Pout ¼ ηA;m ωm þ ηA;m ωm ´ Pth;m 1 :
Ω0 Pth;m
aligning a lensed fiber to a cleaved edge facet to excite the coupling
waveguide. Light is outcoupled from the chip and demultiplexed to detect ð4Þ
the fundamental and second harmonic light separately. b Scanning the
Here ηk;j  κðeÞ
k;j =κk;j
is the cavity-waveguide coupling efficiency
near-infrared laser shows that second-harmonic generation occurs at
for k ∈ {A, B} and j being the index of a specific mode.
wavelengths corresponding to modes of the resonator. c Scanning the
Driving the fundamental frequency ω0 generates light at the
blue pump laser across wavelength shows that many resonances surpass
second harmonic mode Ω0. The efficiency of this process has a
the parametric oscillation threshold. VOA variable optical attenuator,
linear dependence on input power in the low power regime. Once
EDFA erbium-doped fiber amplifier, FPC fiber polarization controller, PM
the additional nonlinear conversion loss experienced by the FH
power meter, OSA optical spectrum analyzer, APD avalanche
mode (proportional to 8∣g0,00A0∣2/κB,0 with zero detuning)
photodiode.
approaches the cavity linewidth κA,0, the cavity’s effective cou-
pling efficiency to the input light is reduced. This leads to a sub-
exceeding 106, which dramatically enhance nonlinear processes linear P−1/3 dependence as the process now converts a substantial
by increasing the lifetimes of the interacting photons. amount of pump photons to second harmonic photons in the
The resonances around the fundamental and second harmo- resonator. A competing oscillation instability leading to para-
nic bands have frequencies ωm and Ωk, with corresponding metric oscillations may prevent observing this power law.
linewidths κA,m and κB,k. We drive with pump frequency nearest At high FH pump powers, the intracavity SH photon popula-
to ω0 and Ω0 in the following experiments. The FH mode fre- tion at Ω0 is large enough to create an instability in the field
quencies vary with index as ωm ≈ ω0 + ζ1m + ζ2m2/2, where ζ1 is amplitude of FH modes Am, causing parametric oscillations when
the free spectral range and ζ2 is a dispersion parameter. Tem- the generated SH intracavity photon number exceeds the
perature tuning of the devices changes the relative detuning threshold condition:
between the modes and gives us fine control over the modal 1
  
2 2 2
detuning μ ≡ Ω0 − 2ω0. The small free spectral range of our B0 ≥ 2δ þ μ  ζ m þ κ κ :
2 2 A;m A;m ð5Þ
device (17.26 GHz and 16.45 GHz at the FH and SH, respec- 16 g 0;mm
tively), allows us to tune μ while keeping the device within a few
degrees of room temperature. We call this a cascaded OPO, since a cascade of two back-to-
The χ(2) optical nonlinearity of the material causes two FH back χ(2) processes leads to parametric oscillation. The threshold
resonances at ωm and ωn, and the SH resonance at Ωk to interact for a cascaded OPO is a function of pump detuning δ, modal
with each other at a rate gk,nm. All of the dynamics of this system detuning μ, and dispersion ζ2m2.
are captured by a set of coupled-mode equations for the funda- We experimentally probe the nonlinear devices with the setup
mental (Am) and second harmonic (Bk) field amplitudes. These presented in Fig. 2a; we use two input paths to drive the resonator
amplitudes correspond to intracavity energies ℏωm∣Am∣2 and with fundamental and second harmonic frequency light—shown
in red and blue, respectively. We use the path connected to a

NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5

a the presented data refers to the on-chip power, accounting for the
Output Power (µW) edge coupling loss.
On-Chip OPO

We study the OPO by driving the device at around 765.8 nm

and recording the generated light at close to twice the wave-
100
length. We temperature tune the modal detuning μ close to zero
to achieve degenerate operation (see Fig. 5 in the "Methods"
Noise floor section). Given the modal detuning’s temperature dependence
10-2
50 100 150 200 300 and our device’s comparatively small free-spectral range (about
On-Chip Pump Power (µW) 17 GHz), we achieve an optimal operating point close to the
b room temperature, at 25.65 ∘C. For the threshold measurement
0
m = -95
we detune μ from zero to allow the most efficient pair of modes
at ω±m to oscillate, following Eq. (3) (see also the condition
45

m = -89
15

OPO Output Power (norm., dB)

m = -79
m = -69 defined by Eq. (16) in the “Methods” section). We plot the
Signal wavelength (nm)

m = -59 power of the generated near infrared light in Fig. 3a. The output
m = -46
-5
m = -36 power vs. input power curve reveals the threshold of oscillation
around 73 μW, which we extract from fitting Eq. (4). A max-
35
15

imum efficiency of 11% is measured. Tuning the pump laser

wavelength allows for effective selection for the frequencies of
m = 36 -10 oscillating signal-idler pairs of modes. By changing the laser
m = 46
m = 59 detuning Δ, we observe seven different OPO wavelength pairs
25

m = 69 generated in the resonator. Figure 3b shows the OPO emission

15

m = 79
m = 89 spectrum as a function of pump detuning with a pump power of
m = 95
c 250 μW. By tuning the pump laser by just 650 MHz, we can
Transmission (norm.)

address signal modes across a band of over 1 THz. Figure 3c

Output Power (µW)

1 2
shows the pump transmission and OPO emitted power as a
On-Chip OPO

function of the pump detuning. Detunings of the pump laser

Pump

0.5 1 relative to the SH cavity mode result in exciting different OPO

modes. We can resolve steps on the transmission and OPO
0 0 emitted power that correspond to switching between different
-500 0 500 operation modes.
Pump Detuning (MHz) To demonstrate second harmonic generation, we drive the FH
mode at 1549.4 nm and measure the resulting frequency doubled
Fig. 3 Optical parametric oscillation. a Threshold characterization of the light at the output. The device temperature is 30.5 ∘C. Figure 4a
OPO, we increase the 765.8 nm laser power until the oscillation begins at a shows the peak SH power generated as a function of input FH
threshold of around 73 μW; above it, the OPO power output follows a power. A maximum efficiency of 12% is achieved with 390 μW of
square-root relation. Dotted line represents detection noise floor. b Tuning input power in the feed waveguide, which agrees with the
of non-degenerate OPO emission with pump wavelength, at a pump power coupled-mode theory (solid lines) that includes only the A0 and
of 250 μW. We can select OPO signal/idler pairs spanning 1537–1545 nm B0 fields. Figure 4b shows how the transmission lineshape and the
(1527–1519 nm) as the pump is swept over 650 MHz. c SH resonance SH response change as a function of pump power. As the pump
lineshape (blue points) aligned with the OPO response (red points) power increases, the transmission lineshape widens and becomes
collected above threshold (250 μW) shows steps corresponding to shallower due to the additional two photon loss induced by the
switching between signal/idler pairs. nonlinearity. At pump powers around 200 μW, the transmission
lineshape forms two distinct valleys, consistent with our coupled-
tunable laser operating in telecommunication wavelengths to mode theory simulations.
study the SHG and cascaded parametric oscillation processes. To At higher input powers (the yellow shaded region of Fig. 4b),
drive a direct OPO, we use the input path connected to the the SH response becomes asymmetrical with a distinct drop in
shorter wavelength laser. The light is coupled into and out of the SH power for negative pump detunings, δ < 0. At these powers,
chip using lensed fibers. We separate the output light using a free- the intracavity SH light is intense enough to create an instability
space setup with a dichroic mirror and send it to Si and InGaAs in the field amplitude of the fundamental modes at ω±m, causing
avalanche photodiodes. We show examples of transmission parametric oscillations as visualized in Fig. 4c. The cascade of
spectra and corresponding SHG and OPO signals in Fig. 2b and c, two χ(2) processes creates the parametric oscillation. The nor-
respectively. We expect to see variation in the efficiency of mal dispersion of the waveguide (ζ2 < 0) creates a lower
second-order processes when probing multiple resonances due to threshold condition for negative pump detunings (δ < 0),
changes in the modal detuning μ across the band for a fixed see Eq. (5). The drop in SH output power at those laser
temperature due to the FSR mismatch between FH and SH and detunings is because SH light at Ω0 converts back to FH power
variations of the quality factors of participating modes. For at ω±m.
spectrally-resolved measurements, we send part of the FH light to We spectrally resolve the cascaded parametric oscillations as a
the optical spectrum analyzer. We calibrate the fiber-to-chip function of laser detuning and confirm that the first sideband
coupling efficiency based on power transmission measurements fundamental modes oscillate at a threshold of 690 μW of on-chip
and fits of theoretical models to nonlinear response data (see pump power. Figure 4d shows multiple sideband oscillations that
“Methods”). The typical edge coupling efficiency across devices occur at a pump power of 930 μW. Particular signal-idler pairs
on the chip is 25–40% at telecom wavelengths and 10–20% for the oscillate as a function of pump laser detuning as expected from
second harmonic depending on fibers and alignment. For each Eq. (5). Disorder in the mode spacing and quality factors causes
experiment, we measure these efficiencies to within less than a certain mode pairs to oscillate before others, consistent with
percent uncertainty (see Table 1 in the "Methods" section). All of coupled-mode simulations.

4 NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5 ARTICLE

Discussion waveguide. We deposit 700 nm of PECVD silicon dioxide at a temperature of

We expect to find ultra-efficient second-order nonlinear photonic 350 ∘C as cladding.
We perform the periodic poling step before waveguide fabrication. For periodic
circuits, such as the frequency doubler and parametric oscillator poling, we use electron-beam evaporated Cr electrodes with an electron beam
demonstrated in this work, in a number of emerging low-power lithography-based liftoff process and apply high-voltage pulses similar to Nagy
and quantum applications in the near future. Together with the et al.44 to invert the crystal domains. Upon completion of the poling, we remove
high-performance integrated devices and components that are the electrodes. We only pole one side of the racetrack resonator, but, in principle,
being developed for the thin-film LN platform, the promise of a both sections could be quasi-phase-matched and increase the nonlinear coupling
rate. This design would require careful control of the relative phase of the
new class of versatile integrated photonic technologies may soon fundamental and second harmonic light. The waveguide we used has normal
be realized. In addition to sources of broadband and quantum dispersion; as a result, the bandwidth of the quasi-phase-matching was
light for sensing and communications, integrated ultra-low-power about 10 nm.
OPOs can be used for computation with coherent Ising Chip edge facet preparation is done using a DISCO DFL7340 laser saw. High
energy pulses are focused into the substrate to create a periodic array of damage
machines40 and cluster states41,42. Note: In the final stages of locations, which act as nucleation sites for crack propagation and result in a
preparing this manuscript we became aware of a demonstration uniform and smooth cleave.
of low-power optical parametric oscillator in a lithium niobate
microresonator43. Experimental setup. We characterize fabricated devices in a simplified
experimental setup shown in Fig. 2a. In the FH input path, we use SMF-28
fibers. 5% of the laser light (Santec TSL-550, 1480–1630 nm) goes into a Mach-
Methods Zehnder interferometer (MZI) with an FSR of 67.7 MHz used to calibrate the
Fabrication. We fabricate all devices with X-cut thin-film lithium niobate-on- relative wavelength during laser wavelength sweeps (not shown in Fig. 2a). 95%
insulator (LNOI) wafers. The material consists of a 500 nm film of LN bonded to a of the light goes to erbium-doped fiber amplifier (EDFA) with a fixed output
2 μm layer of silicon dioxide on top of an LN handle wafer. power of 250 mW followed by a variable optical attenuator. Next, the light
We pattern the optical devices using electron beam lithography (JEOL 6300- passes through a fiber polarization controller (FPC), and we tap 5% of it just
FS, 100-kV) and transfer the design to the LN via Argon ion milling with Ar flow before the input lensed fiber for power calibration with a power meter (Newport
of 15 sccm and a 750 V accelerating voltage. The waveguide width is 1.2 μm, and 918D-IR-OD3R). The light then couples to the chip facet through an SMF-28
the etch depth is 300 nm which leaves a 200 nm slab of LN beneath the lensed fiber.

SH response
1

2 −m +m
10
SHG Power (µW)

Z 0
15
SHG Efficiency (%)

Fund. mode number

100 -10
-9
10 -8

0
10-2 5
Z 8
9
0 10
100 101 102 103
On-chip Pump power (µW) -200 0 200
Detuning (MHz)

Pin = 780 nW 78 µW 123 µW 195 µW 310 µW 490 µW 780 µW 970 µW 1.2 mW

Transmission
1

(norm.)
0

1 SH Power
(norm.)

0
Detuning (250 MHz/division)

Fig. 4 Second harmonic generation and cascaded OPO. a 774.7 nm output power as a function of 1549.4 nm input power (blue circles, left axis) and SHG
efficiency (green diamonds, right axis). Solid lines represent theoretical prediction. Light yellow shading corresponds to the region where we observe
cascaded OPO. Darker shaded region have competing third-order nonlinear processes. b Transmission (top) and SH (bottom) lineshapes evolve as a
function of power. We plot theoretical curves (solid lines) on the top of data (red and blue points) up to the limit where the two-mode model breaks down
and results in a cascaded OPO. A single parameter, the modal detuning μ, is varied by about 0.04κ between fits. c Cascaded OPO scheme—photons at the
fundamental frequency drive the SHG process and create light at 2ω. Sufficiently high power of the SH can drive the parametric oscillation back in the
fundamental frequency range. d Measured cascaded OPO, we observe light generation in modes symmetrically spaced from the pump frequency as the SH
develops asymmetric lineshape shape (top panel).

Table 1 Summary of the measured device parameters.

Device λA (nm) λB (nm) QA (106) QAðiÞ ð106 Þ QB (106) QðiÞ

B ð10 Þ
6
g0,nm (kHz) ηFH (%) ηSH (%)
λA;m ¼ 1521:05 QA;m ¼ 0:68 QðiÞ
A;m ¼ 0:80
OPO 765.77 0.88 1.50 150 37 13
λA;m ¼ 1542:43 QA;m ¼ 0:94
QðiÞ
A;m ¼ 1:50
SHG 1549.40 774.70 0.74 1.2 0.82 1.2 130 26 11

We summarize the wavelengths (λA and λB), total (Q), and internal (Q(i)) quality factors of all of the resonances used in our OPO and SHG experiments. We list the nonlinear coupling factors (g0,nm) and
edge coupling efficiencies at both wavelengths (ηFH and ηSH) for particular devices.

NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5

0 0

65
38

15
15

OPO Output Power (norm., dB)

OPO Output Power (norm., dB)

Signal wavelength (nm)

Signal wavelength (nm)

-5 -5

45
34

15
15

m=0

25
30

-10 -10

15
15

5
26

0
15
15
Transmission (norm.)

Transmission (norm.)
Output Power (µW)

Output Power (µW)

1.5 1
1

On-Chip OPO

On-Chip OPO
0.5
Pump

Pump
0.5 0.75 0 1.5

0.75
0 0
0
-500 0 500 391.75 391.76 391.77 391.78 391.79 391.8
Pump Detuning (MHz) Pump Frequency (THz)

Fig. 5 Tuning between degenerate and nondegenerate optical parametric oscillation. a Tuning of the OPO close to μ = 0, at a pump power of 250 μW.
We observe degenerate parametric oscillation at 1531.7 nm (m = 0) and nondegenerate operation for blue-detuning of the pump laser. b SH resonance
lineshape (blue points) aligned with the OPO response (red points) collected above threshold with 250 μW of on-chip pump power. The distinct feature at
zero-detuning corresponds to degenerate oscillation. c Sweeping the pump laser over four neighboring modes shows that the signal/idler pair center
frequencies are different for each OPO. This is due to the different modal detuning μ = 0 experienced by each OPO due to the difference in dispersion at
765 and 1530 nm. The higher frequency OPOs have larger ∣μ∣ (see discussion in the section “Parametric oscillation theory”). This shows that additional
tuning range of the device's output frequency can be extended to about 2.75 THz by utilizing multiple OPOs of a single resonator while keeping the chip
temperature fixed. d SH resonance lineshape (blue line) aligned with the OPO response (red line) collected above threshold at 250 μW of pump power
on chip.

In the SH path, we use a Velocity TLB-6700 laser that operates in the input edge coupling efficiency to be 13%. We extract the output fiber-to-chip
765–781 nm range. This entire path uses 780HP fiber to maintain single-mode coupling efficiency at the OPO wavelength by fitting the data in Fig. 3a to ηFHPout
operation. A 5% tap outcouples part of the light an MZI with an FSR of 39.9 MHz using Eq. (4). We infer ηFH = 37% coupling efficiency, which we confirm with an
to calibrate laser wavelength sweeps (not shown in Fig. 2a). A variable optical independent transmission measurement.
attenuator controls the remaining laser power, and we control the polarization with
an FPC. 5% of the light goes to a power meter (Newport 918D-SL-OD3R) for input
power calibration, and we focus the rest of it on the chip facet through a 780HP SHG characterization. We characterize the modes contributing to the second
lensed fiber. harmonic generation in an analogous way to the OPO. From the Lorentzian fits at
Once the light exits the output edge facet of the chip, we collect it into a lensed low power we find quality factors of QB,0 = 0.82 × 106, QðiÞ
B;0 ¼ 1:2 ´ 10 , QA,0 =
6
SMF-28 fiber, similar to the one used in the FH input path. We outcouple the light ðiÞ
0.75 × 106, and QA;0 ¼ 1:2 ´ 106 . Moreover, we use a method for fitting nonlinear
into free space and demultiplex with a 1000 nm short pass dichroic mirror. After
the dichroic mirror, SH and FH paths are additionally filtered to ensure no cross lineshapes at high power, as mentioned in the main text. For this purpose, we
talk, and we detect SH and FH light with avalanche photodiodes (Thorlabs solve Eqs. (31) and (30) numerically and fit the resulting curves as a function of
APD410A and Thorlabs APD410, respectively). Variable optical attenuators are detuning to the data. We use the ten lineshapes at the pump power between 80
used before the APDs to avoid saturation. We split 50% of the FH light into an and 620 μW, which allows us to observe changes due to the second-order non-
optical spectrum analyzer (OSA, Yokogawa AQ6370D) for spectrally-resolved linearities but avoid the effects of the cascaded OPO. From this procedure we find
ðiÞ
measurements of the OPO. average QA,0 = 0.74 × 106, and QA;0 ¼ 1:2 ´ 106 and standard deviation of <4%
We use different, but similar, devices on the same chip for the SHG and OPO which agrees with the low power fit. From fitting nonlinear lineshapes, we also
experiments. The chip sits directly on a thermo-electric cooler for temperature extract the coupling rate ∣g0,00∣ to be about 130 kHz, which agrees with our the-
adjustment. oretical prediction of 170 kHz. In the main text, we use averaged values to plot the
theoretical lineshapes and only vary the modal detuning to account for small
temperature fluctuations. For the SHG device, we make transmission measure-
OPO characterization. We characterize all of the optical resonances that take part ments and find the coupling efficiencies to be to be 26% and 11% at the FH and
in the optical parametric oscillation using linear spectroscopy at powers sub- SH, respectively.
stantially below nonlinear effects. For these measurements, we sweep the wave- We calculate the theoretical relationship between the pump power, SHG
length of tunable lasers in the FH and SH bands and fit the transmission dips with power, and SHG efficiency by numerically solving Eqs. (31) and (30) for zero
lorentzian lineshapes. We determine the total and intrinsic quality factors of the detuning. For the solid lines plotted in Fig. 4a, we use quality factors and the
ðiÞ nonlinear coupling rate from the measurements described in the previous
second harmonic mode to be QB,0 = 0.88 × 106 and QB;0 ¼ 1:5 ´ 106 , respectively.
paragraph.
We find the quality factors of the OPO signal modes corresponding to the curve
ðiÞ
in Fig. 2a to be: QA,m = 0.68 × 106, QA;m ¼ 0:8 ´ 106 , QA,−m = 0.94 × 106,
ðiÞ
QA;m ¼ 1:5 ´ 106 . We perform an independent second harmonic generation Resolving OPO lines. We use an OSA (Yokogawa AQ370D) to characterize the
measurements to determine if the FH and SH modes are under or overcoupled. frequency content of the OPO output spectrum as a function of pump laser
The analysis of transmission lineshapes as a function of pump power confirms detuning. With a constant pump power, we repeatedly sweep the laser wavelength
that all modes are undercoupled. From the determined threshold of 73 μW we across the SH resonance and record the SH and FH response with APDs (see
deduce a coupling rate ∣g0,−mm∣ of 150 kHz which is close to the simulated value of section “Experimental setup”). A portion of the generated FH light is detected by
186 kHz. the OSA operating in zero-span mode with a 0.1 nm filter bandwidth, which is less
We measure the input fiber-to-chip coupling with an independent transmission than the ~0.135 nm free spectral range of the FH modes. We step the center
measurement using 780HP lensed fibers at the input and the output chip edges. We wavelength of the OSA across a 40 nm span with a 50% overlap in OSA filter spans.
assume the input and output coupling is identical, an assumption based on We record the detected power on the OSA synchronously with the APD detector
experience with multiple devices on the chip used for the experiment, and find the voltages for each wavelength step. Repeated laser sweeps with different OSA filter

6 NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5 ARTICLE

center wavelengths produce a map of the OPO frequency content as a function of Interestingly, if the modal detuning μ is held constant while the pump
laser detuning shown in Fig. 3b. detuning Δ is swept, the oscillation threshold can select very different modes m
To characterize the cascaded parametric oscillations as shown in Fig. 5d, we first with only small changes in Δ. When the laser is nearly resonant with Ω 0, so Δ is
find every potential OPO line’s precise location (ωm) by performing a broad sweep small compared to the B mode linewidth, the first term in parenthesis in Eq.
of the FH pump laser and record the resonance frequencies. We then proceeded (15) is minimized and does not vary strongly with detuning, while the second
with the measurement in an identical fashion to the standard OPO, but with the term is minimized whenever Δ + μ ≈ ζ2m2. This means that with a fixed laser
0.1 nm wide OSA filters placed precisely at the FH mode locations without any input power, sweeping the laser across the second harmonic mode causes
overlap between filters. oscillation at very different mode numbers and explains the spectrum in
Fig. 3b. For example, if we set Δ ≪ κB,0, we would obtain an approximate
equation for the oscillating mode index (which should be rounded to obtain an
Coupled mode theory equations. The Hamiltonian of the system is used to find
integer, and requires μ + Δ to have the same sign as ζ2):
the equations of motion in the rotating frame.
qffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffi
d κA;m μþΔ
Am ¼  Am  2i ∑ g k;nm An Bk eiδk;nm t  κðeÞ
A;m F m e
iðωL ωm Þt
ð6Þ m ð16Þ
dt 2 kn ζ2
κB;k qffiffiffiffiffiffiffiffi For the real device, we observe disorder in the loss rates for different signal
d
B ¼ B  i ∑ g k;nm Am An eþiδk;nm t  κðeÞ
B;k Gk e
iðΩL Ωk Þt
; ð7Þ modes, which can result from fabrication imperfections or coupler dispersion. We
dt k 2 k mn
can account for that in our threshold calculation
with δk,nm = Ωk − ωn − ωm. Am is the fundamental field amplitude at ωm, and Bk is
the second harmonic field amplitude at Ωk. _Ω0 1 2
Pth;m ¼ ðΔ2 þ ðκB;0 =2Þ2 Þ ´ ððΔ þ μ  ζ 2 m2 Þ þ κA;m κA;m Þ: ð17Þ
16jg 0;mm j2 κðeÞ
B;0
Parametric oscillation theory. We consider the case where the SH modes are To obtain a relation for the OPO power output, we solve Eqs. (6)–(7) for
driven at frequency ΩL and the A modes are not excited and calculate the stability specific modes in steady-state. For the zero detuning of the pump mode Δ = 0 and
criterion for the A modes based on Eq. (7): assuming μ = ζ2m2, we have:
κB;0 qffiffiffiffiffiffiffiffi
d
B ¼ B  κB;0 ðeÞ
G0 eiðΩL Ω0 Þt : ð8Þ 4ig 0;mm Am B0
dt 0 2 0 Am ¼ ð18Þ
κA;m
~
We go into a rotating frame B0 ¼ B0 e iΔt
with frequency Δ = ΩL − Ω0 defined
as the detuning between the laser drive and the B mode, which we can solve in qffiffiffiffiffiffiffiffi
ðeÞ
steady-state to obtain: 2ig 0;mm Am Am þ κB;0 G0
qffiffiffiffiffiffiffiffi B0 ¼ : ð19Þ
κB;0 =2
κðeÞ
B;0 G0
~
B0 ¼ : ð9Þ Now, if we note that the oscillating amplitudes and coupling rate are complex
iΔ  κB;0 =2
Am ¼ jAm j expðiθm Þ, g k;00 ¼ jg k;00 j expðiφÞ, we can substitute Eq. (18) to (19) and
We now consider two A modes at frequencies ωm and ω−m which are coupled obtain
by the intracavity population of B0. Their coupling leads to a pair of equations qffiffiffiffiffiffiffiffi
ðeÞ
d κA;m κA;m κB;0 jAm j κB;0
A ¼ A  2ig 0;mm Am B0 eiδ0;mm t  2ijAm jjAm j þ G eiðφθm θm Þ ¼ 0: ð20Þ
dt m 2 m 8ijg 0;mm j jAm j
2
jg 0;mm j 0
d  κA;m 
A ¼ Am þ 2ig 0;mm Am B0 eiδ0;mm t This requires the exponential expðiðφ  θm  θm ÞÞ to be purely imaginary,
dt m 2
φ − θm − θ−m = π/2 + d ⋅ π, where d 2 Z. This phase relation shows that the
which become unstable for sufficiently large ∣B0∣. To see this note that sum of the phases of the OPO output are locked to the phase of the pump. As a
δ0,−mm = (Ω0 − 2ω0) − ζ2m2 allowing us to move into a rotating frame with result, we can use Eq. (18) to find that
  sffiffiffiffiffiffiffiffiffiffiffi
~ m exp i Δ þ μ  ζ 2 m t
2
Am ¼ A ð10Þ jAm j κA;m
2 ¼ ; ð21Þ
jAm j κA;m
where μ ≡ Ω0 − 2ω0 is the modal detuning between the driven SH and closest FH
mode, which in our experiment is set by tuning the temperature. In this frame, the and solve Eq. (20) for the photon flux of both signal modes of the OPO:
equations become time-independent, and we obtain the stability criterion qffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffiffiffiffi ðeÞ
(assuming κ±m are equal for simplicity): κA;m κB;0 G0 κB;0 κA;m
Am ¼
2
 ð22Þ
2
16jg 0;mm j2 jB0 j2 ≥ ðΔ þ μ  ζ 2 m2 Þ þ ðκA;m Þ2 : ð11Þ κA;m 2jg 0;mm j 16jg 0;mm j2
To relate this to the input photon flux at the SH frequency Ω0, we replace B0 qffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffiffiffiffi
using Eq. (9), to obtain
κA;m κðeÞ
B;0 G0 κB;0 κA;m
ðeÞ 2 Am
2
¼  : ð23Þ
16jg 0;mm j2 jG0 j2 κB;0 ≥ ðΔ2 þ ðκB;0 =2Þ2 Þ ´ ððΔ þ μ  ζ 2 m2 Þ þ ðκA;m Þ2 Þ: ð12Þ κA;m 2jg 0;mm j 16jg 0;mm j2
We can see from here that the lowest degenerate oscillation threshold can be To analyze the total output power of the OPO in experiment, we sum over the
achieved when μ = 0 and Δ = 0: power of two signal modes
4jg 0;mm j2 jG0 j2 κðeÞ 2 2
ð13Þ sffiffiffiffiffiffiffiffiffiffi !
B;0 ≥ ðκB;0 =2Þ ðκA;m =2Þ or; 4ηB;0 PB;0
Pout ¼ ðηA;m ωm þ ηA;m ωm Þ ´ Pth;m 1 ; ð24Þ
Ω0 Pth;m
_Ω0 κ2B;0 κ2A;0
Pth;0 ¼ : ð14Þ
64jg 0;00 j2 κðeÞ
B;0
ðeÞ
with ηk ¼ κk;0 =κk;0 for k = A, B being the cavity-waveguide coupling efficiency.
More generally, the OPO will oscillate first in the mode m for which Pth,m is the PB,0 is the pump power of the SH mode and Pth,m is a generalized OPO
lowest, where threshold, which includes disorder in the total loss rates of fundamental
modes:
_Ω0 1 2
Pth;m ¼ ðΔ2 þ ðκB;0 =2Þ2 Þ ´ ððΔ þ μ  ζ 2 m2 Þ þ ðκA;m Þ2 Þ: ð15Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
16jg 0;mm j2 κðeÞ _Ω0 κ2B;0 κA;m κA;m _Ω0 κA;m κA;m
B;0 Pth;m ¼ ðeÞ
¼ ; ð25Þ
Here we’ve assumed again that the losses for the ±m modes are equal. 64jg 0;mm j2
κB;0 16ηB C0;m
Equation (15) shows that we can use the modal detuning μ and the driving pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
detuning Δ to select which modes reach threshold first and oscillate as the where C0;m  4jg 0;mm j2 =ð κA;m κA;m κB;0 Þ is the vacuum cooperativity for m th
power is increased. Assuming that g0,−mm does not change significantly with the pair of signal modes. Note that this relation agrees with Eq. (17) for the case of
mode number, we see that for on-resonant driving Δ = 0, a minimum threshold modal and laser detuning optimized for m th OPO sideband.
can be achieved when μ = ζ2m2, as long as μ and ζ2 have the same sign. In our
case, the waveguide has normal dispersion, sop ζ2ffiffiffiffiffiffiffiffiffi
is negative, and we have Second-harmonic generation efficiency. Starting from the coupled mode Eqs. (6)
roughly ζ2/2π = −100 kHz. The relation m  μ=ζ 2 shows that the mode and (7), we now assume that only A0 is excited, i.e., we are driving the mode at ω0
number selected is very sensitive to the modal detuning (set by temperature) and all other mode FH amplitudes are 0:
which makes the degenerate oscillation mode challenging to obtain in a system
κA;0 qffiffiffiffiffiffiffiffi
with a large resonator and therefore very small ζ2 mode-spacing dispersion d
A ¼ A  2i ∑ g k;00 A0 Bk eiδk;00 t  κðeÞ
A;0 F 0 e
iðωL ω0 Þt
ð26Þ
parameter. dt 0 2 0 k

NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5

this model to break down and the system to go into cascaded optical parametric
κB;k oscillation.
d
B ¼ B  ig k;00 A20 eþiδk;00 t : ð27Þ
dt k 2 k
To solve these equations, we go into a frame that rotates with the laser detuning Cascaded optical parametric oscillation. Consider the same driving as in the
frequency δ = ωL − ω0, so A0 ¼ A ~ 0 eiδt , Bk ¼ B
~ k eið2δδk;00 Þt ¼ B ~ k eið2ωL Ωk Þt : previous section, where a laser drive at the fundamental with frequency ωL excites
~ 0 and generates an intracavity population in the second harmonic mode B
A ~ 0 . From

κA;0  q ffiffiffiffiffiffiffi
ffi
d ~ ~ 0  2i ∑ g A ~ ~ the section on the oscillation threshold, we know that at a sufficiently value of jB ~ 0 j,
A0 ¼ iδ  A k;00 0 Bk  κðeÞ
A;0 F 0 ð28Þ
dt 2 k the equations of motion for mode amplitudes A ~ ± m become unstable and set of

oscillations, with a threshold condition given by an equation very similar to
d~ κB;k 
Bk ¼ ið2ωL  Ωk Þ  B ~2:
~ k  ig k;00 A ð29Þ Eq. (11):
0
dt 2  
~ 0 2 ≥ 2δ þ μ  ζ 2 m2 2 þ κA;m κA;m :
16jg 0;mm j2 B
We can solve these in steady state to obtain:
We call this a cascaded OPO, since a cascade of two back-to-back χ(2)
~
ig k;00 A
2
processes lead to parametric oscillation. It is clear from the oscillation condition
~k ¼
B
0
κB;k ð30Þ
ið2ωL  Ωk Þ  2
that the threshold is highly detuning-dependent, and also depends on the
dispersion parameter ζ2. In our case, ζ2/2π ≈ −100 kHz and so the oscillation

qffiffiffiffiffiffiffiffi threshold is lower with the laser tuned to the red side (δ < 0) when the modal
κA;0  ~ 0 j2
2jg k;00 j2 jA detuning μ ≈ 0.
0 ¼ iδ  ~0 þ ∑
A ~
κB;k A0 
ðeÞ
κA;0 F0 ð31Þ
2 k ið2ωL  Ωk Þ 
2

There are a couple of interesting things to note about the last equation. Note that Nonlinear coupling rate. We derive the nonlinear coupling rate from the inter-
each SH mode at Ωk contributes effective nonlinear loss and detuning terms to the action energy density in the three-wave mixing process. Given the electric field
FH mode at ω0: distribution E = (Ex, Ey, Ez) The interaction energy density is given by:
2ð2ωL  Ωk Þ ε ð2Þ α β γ
Detuning:  ∑ 2 2 U χð2Þ ¼ 0 ∑ χ αβγ E E E ; ð38Þ
κB;k 2 g k;00 A0 ð32Þ 3 αβγ
k ð2ωL  Ωk Þ2 þ 2
each of the three waves can be expressed using spatial complex amplitudes Em, En,
2κB;k 2 Ek as follows:
Loss: ∑ 2
κB;k 2 g k;00 A0 ð33Þ
k ð2ωL  Ωk Þ2 þ 2 E ¼ Am Em eiωm t þ An En eiωn t þ Bk Ek eiΩk t þ h:c: ð39Þ
For large 2ωL − Ωk ≫ κb, we see an effect which is primarily a frequency shift and To calculate the nonlinear coupling rate we focus on three specific modes in the
looks much like a χ(3) cavity frequency shift. sum and evaluate Eq. (38) by averaging away the rapidly rotating terms. It selects
From here on, we assume that only one SH mode (k = 0) is significantly excited. only energy-conserving terms of the sum. Since the second-order nonlinear tensor
The photons generated at the B0 mode frequency are emitted from the device
qffiffiffiffiffiffiffi
ffi has a full permutation symmetry, for the non-degenerate we find that
generating a photon flux ∣Gout,0∣2 at the SH frequency where Gout;0 ¼ κðeÞ ~ ð2Þ  α β γ  
B;0 B0 . To U k;nm ¼ 2ε0 ∑ χ αβγ Ek Em En Bk Am An þ h:c: ¼ 2ε0 ðEk  χ ð2Þ : Em En Bk Am An þ h:c:Þ:
χ ð2Þ
find B ~ 0 (Eq. (30)), which is given implicitly by
~ 0 , we need to calculate A αβγ

ð40Þ

κA;0  ~0 2
2jg 0;00 j2 A qffiffiffiffiffiffiffiffi
0 ¼ iδ  ~0 þ ~
κB;0 A0  κðeÞ ð34Þ Integrating over this energy density gives us the total energy of the system, which
A A;0 F 0 :
2 ið2ωL  Ω0 Þ  2 we use to derive the equations of motion (6) and (7). We choose normalization of
For the fits shown in the paper, this equation was solved numerically. Here we the modal field Ek so that the total energy corresponding to an amplitude Ak is
assume δ = 0 and approximate the solutions in two limits, (1) the low-power limit ℏωk∣Ak∣2. More precisely, given unitless field profiles ei (with maxðei Þ ¼ 1), we
where the first term is dominant, and (2) the high-power limit where the second introduce normalization factors Ni, defined by Ei = Niei. The energy condition then
term is dominant. The cross-over between these two limits occurs at fixes these normalization factors as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2C0 nAð0Þ ¼ 1 ð35Þ
R
_ωi
Ni ¼
2 ei εðrÞei dV
where C0 = 4∣g0,00 A,0κB,0 is a cooperativity parameter and
∣2/κ ¼ nAð0Þ ðeÞ
4κA;0 jF 0 j2 =κ2A;0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
is the number of intracavity photons which would be excited in the absence of _ωi
¼ R ð41Þ
nonlinearity. Solving the above equation in the two limits gives us 2ε0 L ei εr ðrÞei dA
! sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð0Þ 1=3
~ 0 2 ¼ nA
~ 0 2 ¼ nAð0Þ ; and A _ωi 1
A ¼ pffiffiffiffiffiffi
4C20 2ε0 L n2i Ai
R
for the low- and high-power limits, respectively. We define the second harmonic Here, we introduced the effective mode
R area for each mode as Ai ¼ A jei j2 dA, and
generation power efficiency define the average index as n 2i ¼ ei εr ðrÞei dA=Ai . To find the energy, we inte-
2 grate Eq. (40) over the mode volume. We account for a partially-poled racetrack
Pout 2 Gout;0 resonator by introducing the poled length fraction λ as a ratio of the poled region
ηSHG  ¼ 2 ; ð36Þ
Pin F0 to the total resonator length L. The final expression for the nonlinear coupling rate
is given by:
~ 0 j2 :
which after some manipulation, can be written in terms of jA sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
8ηA ηB C0 ~ 4 λ _ωm ωn Ωk O
g k;nm ¼ pffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð42Þ
ηSHG ¼ A0 2π ε0 L n2k n2m n2n Am An Ak
nð0Þ
8 A
< 8ηA ηB C 0 nAð0Þ low power ð37Þ where O represents the mode overlap integral over the waveguide cross-section area
¼ 4ηA ηB
Z
: high power O¼ ek  χ ð2Þ : em en dA: ð43Þ
ð2nAð0Þ C0 Þ
1=3

A
It is apparent that at low power, the efficiency increases linearly, but is then For our numerical waveguide calculations we use a finite-element mode solver
saturated at high power. This can be understood from an impedance matching (COMSOL).
perspective. As the pump power is increased, the FH cavity resonance senses a two-
~ 0 j2 =κB;0 (see Eq. (33)). As this loss starts to
photon loss proportional to 8jg 0;00 A Numerical simulations of dynamics. We numerically integrate the coupled-mode
exceed the cavity linewidth, its effective coupling rate to the waveguide is reduced, differential Eqs. (6) and (7) to understand how the transmission spectra change when
preventing input light from coupling efficiently into the cavity to be frequency- the system starts to oscillate and how disorder affects the emission spectra of the
doubled. Designing an overcoupled resonator can compensate for the nonlinear cascaded OPO. We integrate the coupled-mode equations with 181 A modes and 31 B
loss rate and allow for higher maximum efficiencies compared to critically coupled modes for 600 ns which is sufficiently long for the system to stabilize. We use the
or undercoupled resonators, at the expense of increased OPO threshold power. At measured parameters from the SHG experiment for the ω0 and Ω0 modes, and assume
very high power, the efficiency actually begins to go down as P−1/3. The model that the other modes are spaced by the measured FSR (which agrees with the theory
assumes that only the B ~ 0 modes are excited. As we saw in the case of a
~ 0 and A prediction) and have the same quality factors. The resulting spectra for are shown in
directly driven OPO, at sufficiently large B ~ ± m start to oscillate, which causes
~0 , A Fig. 6. We then perform the same simulation but with the quality factors and

8 NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5 ARTICLE

a Output Power (norm., dB) c

-15 -10 -5 0 Pa = 0.44 µW 78 µW 120 µW 200 µW 310 µW 490 µW 780 µW 0.97 mW 1.2 mW
1

Transmission
(norm.)
50
Mode Number
Fundamental

0
0

-50
b 1

SH Power
(norm.)
SH Response

1
(norm.)

0.5

0 0
-400 -200 0 200 400
Detuning (MHz) Detuning (250 MHz/division)

Fig. 6 Simulated effect of 0.1% disorder in the mode-to-mode loss rates on the cascaded OPO. a Cascaded OPO signal as a function of pump detuning,
simulated at 0.97 mW of pump power. b Second Harmonic lineshape corresponding to the cascaded OPO in panel (a). c Lineshape evolution as a function
of power, asymmetry is developed below 780 μW, which agrees with the experiment.

a Output Power (norm., dB) c

-15 -10 -5 0 Pa = 0.44 µW 78 µW 120 µW 200 µW 310 µW 490 µW 780 µW 0.97 mW 1.2 mW
1

Transmission
(norm.)
50
Mode Number
Fundamental

0
0

-50
b 1

SH Power
(norm.)
SH Response

1
(norm.)

0.5

0 0
-400 -200 0 200 400
Detuning (MHz) Detuning (250 MHz/division)

Fig. 7 Simulated effect of 1% disorder in the mode-to-mode loss rates on the cascaded OPO. a Cascaded OPO signal as a function of pump detuning,
simulated at 0.97 mW of pump power. b Second Harmonic lineshape corresponding to the cascaded OPO in panel (a). c Lineshape evolution as a function
of power, asymmetry is developed below 780 μW, which agrees with the experiment.

a Output Power (norm., dB) c

-15 -10 -5 0 Pa = 0.44 µW 78 µW 120 µW 200 µW 310 µW 490 µW 780 µW 0.97 mW 1.2 mW
1
Transmission
(norm.)

50
Mode Number
Fundamental

0
0

-50
b 1
SH Power
(norm.)
SH Response

1
(norm.)

0.5

0 0
-400 -200 0 200 400
Detuning (MHz) Detuning (250 MHz/division)

Fig. 8 Simulated effect of 10% disorder in the mode-to-mode loss rates on the cascaded OPO. a Cascaded OPO signal as a function of pump detuning,
simulated at 0.97 mW of pump power. b Second Harmonic lineshape corresponding to the cascaded OPO in panel (a). c Lineshape evolution as a function
of power, asymmetry is developed around 490 μW, lower than in the experiment.

NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5

detunings of the other modes now having disorder (normally distributed fluctuations 29. He, Y. et al. Self-starting bi-chromatic LiNbO3 soliton microcomb. Optica 6,
of total Q and mode frequency) on the order of 1% (Fig. 7) and 10% (Fig. 8) of the 1138 (2019).
cavity linewidth. 30. Ilchenko, V. S., Savchenkov, A. A., Matsko, A. B. & Maleki, L. Nonlinear
optics and crystalline whispering gallery mode cavities. Phys. Rev. Lett. 92,
043903 (2004).
31. Beckmann, T. et al. Highly tunable low-threshold optical parametric
Data availability oscillation in radially poled whispering gallery resonators. Phys. Rev. Lett. 106,
The data sets generated during and/or analyzed during this study are available from the 143903 (2011).
corresponding authors on request. 32. McKenna, T. P. et al. Cryogenic microwave-to-optical conversion using a triply
resonant lithium-niobate-on-sapphire transducer. Optica 7, 1737 (2020).
Received: 26 May 2021; Accepted: 6 June 2022; 33. Holzgrafe, J. et al. Cavity electrooptics in thin-film lithium niobate for efficient
microwave-to-optical transduction. Optica 7, 1714 (2020).
34. Xu, Y. et al. Bidirectional interconversion of microwave and light with thin-
film lithium niobate. Nat. Commun. 12, 4453 (2021).
35. Luo, R. et al. On-chip second-harmonic generation and broadband parametric
References down-conversion in a lithium niobate microresonator. Opt. Express 25, 24531
1. Jalali, B. & Fathpour, S. Silicon photonics. J. Light. Technol. 24, 4600 (2006). (2017).
2. Absil, P. P. et al. Silicon photonics integrated circuits: a manufacturing platform 36. Zhao, J., Ma, C., Rüsing, M. & Mookherjea, S. High quality entangled photon
for high density, low power optical I/O’s. Opt. Express 23, 9369 (2015). pair generation in periodically poled thin-film lithium niobate waveguides.
3. Moss, D. J., Morandotti, R., Gaeta, A. L. & Lipson, M. New CMOS-compatible Phys. Rev. Lett. 124, 163603 (2020).
platforms based on silicon nitride and Hydex for nonlinear optics. Nat. 37. Furukawa, Y., Kitamura, K., Takekawa, S., Niwa, K. & Hatano, H.
Photon. 7, 597 (2013). Stoichiometric Mg:LiNbO3 as an effective material for nonlinear optics. Opt.
4. Romero-García, S., Merget, F., Zhong, F., Finkelstein, H. & Witzens, J. Silicon Lett. 23, 1892 (1998).
nitride CMOS-compatible platform for integrated photonics applications at 38. Rüsing, M., Zhao, J. & Mookherjea, S. Second harmonic microscopy of poled
visible wavelengths. Opt. Express 21, 14036 (2013). x-cut thin film lithium niobate: Understanding the contrast mechanism. J.
5. Rahim, A. et al. Expanding the silicon photonics portfolio with silicon nitride Appl. Phys. 126, 114105 (2019).
photonic integrated circuits. J. Light. Technol. 35, 639 (2017). 39. Moille, G. et al. Broadband resonator-waveguide coupling for efficient
6. Gaeta, A. L., Lipson, M. & Kippenberg, T. J. Photonic-chip-based frequency extraction of octave-spanning microcombs. Opt. Lett. 44, 4737 (2019).
combs. Nat. Photon. 13, 158 (2019). 40. Yamamoto, Y. et al. Coherent Ising machines–optical neural networks
7. Shen, B. et al. Integrated turnkey soliton microcombs. Nature 582, 365 (2020). operating at the quantum limit. npj Quant. Inf. 3, 49 (2017).
8. Okawachi, Y. et al. Octave-spanning frequency comb generation in a silicon 41. Yokoyama, S. et al. Ultra-large-scale continuous-variable cluster states
nitride chip. Opt. Lett. 36, 3398 (2011). multiplexed in the time domain. Nat. Photon. 7, 982 (2013).
9. Li, Q., Davanço, M. & Srinivasan, K. Efficient and low-noise single-photon- 42. Chen, M., Menicucci, N. C. & Pfister, O. Experimental realization of
level frequency conversion interfaces using silicon nanophotonics. Nat. multipartite entanglement of 60 modes of a quantum optical frequency comb.
Photon. 10, 406 (2016). Phys. Rev. Lett. 112, 120505 (2014).
10. Zhao, Y. et al. Near-degenerate quadrature-squeezed vacuum generation on a 43. Lu, J. et al. Ultralow-threshold thin-film lithium niobate optical parametric
silicon-nitride chip. Phys. Rev. Lett. 124, 193601 (2020). oscillator. Optica 8, 539–544 (2021).
11. Vaidya, V. D. et al. Broadband quadrature-squeezed vacuum and nonclassical 44. Nagy, J. T. & Reano, R. M. Reducing leakage current during periodic poling of ion-
photon number correlations from a nanophotonic device. Sci. Adv. 6, sliced x-cut MgO doped lithium niobate thin films. Opt. Mater. Exp. 9, 3146
eaba9186 (2020). (2019).
12. Timurdogan, E., Poulton, C. V., Byrd, M. J. & Watts, M. R. Electric field-
induced second-order nonlinear optical effects in silicon waveguides. Nat.
Photon. 11, 200 (2017). Acknowledgements
13. Cazzanelli, M. & Schilling, J. Second order optical nonlinearity in silicon by The authors wish to thank NTT Research for their financial and technical support. This
symmetry breaking. Appl. Phys. Rev. 3, 011104 (2016). work was funded by the U.S. Department of Defense through the DARPA Young Faculty
14. Chang, L. et al. Heterogeneous integration of lithium niobate and silicon Award (YFA), the DARPA LUMOS program (both supported by Dr. Gordon Keeler), and
nitride waveguides for wafer-scale photonic integrated circuits on silicon. Opt. through the U.S. Department of Energy through Grant No. DE-AC02-76SF00515 (through
Lett. 42, 803 (2017). SLAC). Part of this work was performed at the Stanford Nano Shared Facilities (SNSF),
15. Weigel, P. O. et al. Lightwave circuits in lithium niobate through hybrid supported by the U.S. National Science Foundation under award ECCS-2026822. H.S.S.
waveguides with silicon photonics. Sci. Rep. 6, 22301 (2016). acknowledges support from the Urbanek Family Fellowship. J.F.H. was supported by the
16. Bruch, A. W., Liu, X., Surya, J. B., Zou, C.-L. & Tang, H. X. On-chip χ2 National Science Foundation Graduate Research Fellowship Program. V.A. was supported
microring optical parametric oscillator. Optica 6, 1361 (2019). by the Stanford Q-FARM Bloch Fellowship Program. A.-H.S.N. acknowledges the David
17. Mackwitz, P. et al. Periodic domain inversion in x-cut single crystal lithium and Lucille Packard Fellowship, and the Stanford University Terman Fellowship.
niobate thin film. Appl. Phys. Lett. 108, 152902 (2016).
18. Chang, L. et al. Thin film wavelength converters for photonic integrated
circuits. Optica 3, 531 (2016). Author contributions
19. Jankowski, M. et al. Ultrabroadband nonlinear optics in nanophotonic T.P.M. and H.S.S. designed the device. T.P.M., H.S.S., and V.A. fabricated the device.
periodically poled lithium niobate waveguides. Optica 7, 40 (2020). T.P.M., H.S.S., V.A., J.M., C.J.S., J.F.H., and C.L. developed the fabrication process. V.A.,
20. Wang, C. et al. Ultrahigh-efficiency wavelength conversion in nanophotonic M.J., M.M.F., and A.H.S.-N. provided experimental and theoretical support. T.P.M. and
periodically poled lithium niobate waveguides. Optica 5, 1438 (2018). H.S.S. performed the experiments and analyzed the data. A.H.S.-N. performed numerical
21. Lu, J., Li, M., Zou, C.-L., Al Sayem, A. & Tang, H. X. Toward 1% single- simulations. T.P.M., H.S.S., and A.H.S.-N. wrote the manuscript. T.P.M., H.S.S., and
photon anharmonicity with periodically poled lithium niobate microring A.H.S.-N. conceived the experiment, and A.H.S.-N. supervised all efforts.
resonators. Optica 7, 1654 (2020).
22. Rao, A. et al. Actively-monitored periodic-poling in thin-film lithium niobate Competing interests
photonic waveguides with ultrahigh nonlinear conversion efficiency of 4600% The authors declare no competing interests.
W 1 cm2. Opt. Express 27, 25920 (2019).
23. Chen, J.-Y. et al. Efficient and highly tunable second-harmonic generation in Z-cut
periodically poled lithium niobate nanowaveguides. Opt. Lett. 45, 3789 (2020). Additional information
24. Zhang, M., Wang, C., Cheng, R., Shams-Ansari, A. & Lončar, M. Monolithic Correspondence and requests for materials should be addressed to Amir H. Safavi-
ultra-high-Q lithium niobate microring resonator. Optica 4, 1536 (2017). Naeini.
25. Wang, C. et al. Integrated lithium niobate electro-optic modulators operating
at CMOS-compatible voltages. Nature 562, 101 (2018). Peer review information Nature Communications thanks Lin Chang and the other
26. Zhang, M. et al. Broadband electrooptic frequency comb generation in a anonymous reviewer(s) for their contribution to the peer review of this work.
lithium niobate microring resonator. Nature 568, 373 (2019).
27. Li, M. et al. Lithium niobate photoniccrystal electro-optic modulator. Nat. Reprints and permission information is available at http://www.nature.com/reprints
Commun. 11, 4123 (2020).
28. Wang, C., Zhang, M., Yu, M., Zhu, R., Hu, H. & Loncar, M. Monolithic Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
lithium niobate photonic circuits for Kerr frequency comb generation and published maps and institutional affiliations.
modulation. Nat. Commun. 10, 978 (2019).

10 NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31134-5 ARTICLE

Open Access This article is licensed under a Creative Commons

Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit http://creativecommons.org/
licenses/by/4.0/.

© The Author(s) 2022

NATURE COMMUNICATIONS | (2022)13:4532 | https://doi.org/10.1038/s41467-022-31134-5 | www.nature.com/naturecommunications 11