Systematic Investigation of Millimeter-Wave Optic Modulation

Performance in Thin-Film Lithium Niobate

Yiwen Zhang,1 Linbo Shao,2,3 Jingwei Yang,1 Zhaoxi Chen,1 Ke Zhang,1 Kam-Man Shum,4 Di Zhu,
2,5,6
Chi Hou Chan, 1,4 Marko Lončar, 2 and Cheng Wang1,4,*
1Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong, China
2JohnA. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts
02138, USA
3Bradley Department of Electrical and Computer Engineering, Virginia Tech, 1185 Perry Street, Blacksburg,
Virginia 24061, USA
4StateKey Laboratory of Terahertz and Millimeter Waves, City University of Hong Kong, Kowloon, Hong
Kong, China
5Institute
of Materials Research and Engineering, Agency for Science, Technology and Research
(A*STAR), Singapore, 138634, Singapore
6
[email protected]
*[email protected]

Abstract: Millimeter-wave (mmWave) band (30 - 300 GHz) is an emerging spectrum range for
wireless communication, short-range radar and sensor applications. mmWave-optic modulators
that could efficiently convert mmWave signals into optical domain are crucial components for
long-haul transmission of mmWave signals through optical networks. At these ultrahigh
frequencies, however, the modulation performances are highly sensitive to the transmission line loss
as well as the velocity- and impedance-matching conditions, while precise measurements and
modeling of these parameters are often non-trivial. Here we present a systematic investigation of
the mmWave-optic modulation performances of thin-film lithium niobate modulators through
theoretical modeling, electrical verifications and electro-optic measurements at frequencies
up to 325 GHz. Based on our experimentally verified model, we demonstrate thin-film
lithium niobate mmWave-optic modulators with a measured 3-dB electro-optic bandwidth of
170 GHz and a 6-dB bandwidth of 295 GHz. The device also shows a low RF half-wave
voltage of 7.3 V measured at an ultrahigh modulation frequency of 250 GHz. This work
provides a comprehensive guideline for the design and characterization of mmWave-optic
modulators and paves the way toward future integrated mmWave photonic systems for
beyond-5G communication and radar applications.
1. Introduction
For almost all mmWave-photonic systems, a key
Recent years have witnessed a rapid growth of global component is an electro-optic modulator that could
wireless network traffic. To keep up with the demand of efficiently convert mmWave signals into the optical
the ever-increasing data capacity, it is an attractive and domain [3, 4]. However, it is a highly non-trivial task for
natural solution to explore new spectral bands that are electro-optic modulators to operate at high mmWave
less congested than currently used microwave bands. frequencies, in particular the range between 100 - 300
The millimeter wave (mmWave) band (i.e., 30 - 300 GHz. The electro-optic bandwidths of semiconductor-
GHz) is particularly interesting since it has higher based (e.g., Si and InP) modulators could barely reach
frequencies than microwave bands and therefore allows this range, limited by their carrier lifetimes [5, 6]. Other
for much larger channel bandwidths according to emerging material platforms like electro-optic polymer
Shannon’s theorem. mmWaves are also important for or graphene have shown modulation bandwidths into the
short-range radar and sensing applications [1]. However, terahertz regime [7, 8], yet their long-term stability and
the transmission and processing of mmWave signals is scalability remain to be proven. Lithium niobate (LN,
challenging (and costly) due to exacerbating metallic LiNbO3) is a promising candidate for mmWave-optic
losses as well as the gain-bandwidth trade-off of modulation purposes, since the Pockels-based electro-
traditional electronic components at high frequencies. optic effect in LN intrinsically happens on femtosecond
mmWave photonics is a promising solution to address timescales and the material itself has been industry
these issues in a cost-effective manner, since it allows proven for decades [9]. However, the bandwidths of
generation, transmission and processing of mmWave traditional off-the-shelf LN modulators are typically
signals in the low-loss optical domain, in analogy to the limited to < 35 GHz, since the weak electro-optic
concept of microwave photonics [2]. interactions in ion-diffused waveguides lead to the
requirement of long modulation electrodes and large RF design guidelines, we experimentally demonstrate a 5.8
losses, especially at high frequencies. mm long Mach-Zehnder modulator on TFLN with 3-dB
and 6-dB electro-optic bandwidths of 175 GHz and 295
Benefiting from the development of ion-slicing and
GHz, respectively (with reference to 1 GHz). The
nanofabrication technologies, the thin-film LN (TFLN)
measured RF Vπ is as low as 7.3 V at an ultrahigh
platform has recently shown great promise for pushing
frequency of 250 GHz nearly at the upper bound of the
the operation bandwidths of LN modulators into the
mmWave spectrum, showing practical relevance for the
mmWave regime while exhibiting smaller device
majority of mmWave photonics applications.
footprints and lower power consumption [9]. Owing to
(a)
the much larger refractive index contrast and better Antenna
confined optical mode in TFLN, the modulation
mmWave signal

electrodes could be placed much closer to the optical Carrier

Laser
mmWave-optical
modulator
Optical signal
processor

waveguide, leading to substantially increased electro- (b)

optic overlap and shorter modulation electrodes required. G Optical sig
signal G
As a result, a number of high-performance TFLN mmWave input
Optical input
mmWave output
Optical output
S S
modulators have been developed, demonstrating low mmWave signal

half-wave voltages (Vπ) [10, 11], high modulation x z

y
G G

bandwidths around 100 GHz [12-17], as well as

(c) (d)
ultrahigh linearity [18]. The high-performance TFLN Air

SiO
1

modulators could potentially be further integrated with Cu Cu

frequency comb sources [19, 20], tunable filters [21, 22] LN

and low-loss delay lines [23] on the same platform for x SiO

future microwave- and mmWave-photonic applications. z 0

(e)
While many of these demonstrated modulators have
theoretically predicted modulation capabilities much
beyond 100 GHz [14-16], the experimental 100 µm
demonstrations are limited. In particular, it remains
Fig. 1. (a) Schematic illustration of a future mmWave-photonic system,
unclear whether the performances would match at the heart of which sits the mmWave-optic modulator that converts
theoretical predictions due to the lack of electrical and mmWave signals into the optical domain. (b) Schematic of the TFLN
electro-optic measurements at ultrahigh frequencies. In mmWave-optic modulator, where velocity matching between the
this range, the electro-optic efficiency is very sensitive optical and mmWave signals, impedance matching and RF loss
conditions determine the ultimately achievable modulation bandwidths.
to velocity-matching condition and RF losses, and any (c) Simulated optical mode profile (Ez) in the TFLN rib waveguide. (d)
deviation from the theoretical models can negatively Simulated mmWave profile (Ez) at a frequency of 300 GHz. (e) Optical
impact the modulator performance. For example, micrograph of a fabricated device (the darker regions at the two ends
Mercante et al. have measured the electro-optic are exposed areas for electrical contacts, whereas other parts of the chip
are cladded with silicon dioxide).
responses of TFLN modulators at frequencies up to 500
GHz showing good potential, but the measured electro- 2. mmWave-optic modulator design methodology
optic response substantially differs from the theoretically
predicted curve since electrical measurements (and in Figure 1(a) illustrates an envisioned a future mmWave-
turn, model verifications) are limited to 110 GHz [15]. photonic system, where mmWave signals received by an
The measured 3-dB electro-optic bandwidth of the antenna could be efficiently converted into optical
device is also limited to ~ 40 GHz. The lack of a domian via a modulator, filtered and processed using
comprehensive design guideline with electrical and optical techniques, and/or transmitted over long
electro-optic measurements-based verifications has distances through optical fiber links. An electro-optic
become a major hurdle for TFLN modulators toward modulator with a broad bandwidth covering the entire
future mmWave-photonic applications. mmWave range is vital of such a system. Importantly,
the peripheral components, including antennas [3], lasers
This work provides a systematic investigation of the [24], optical filters [21, 22], amplifiers [24, 25] and delay
velocity- and impedance-matching conditions and RF lines [23], have all been demonstrated on the LN
losses of TFLN modulators based on electrical and platform and could potentially allow the integration of
electro-optic measurements at frequencies up to 325 this entire system on the same chip.
GHz. Our results show excellent agreement not only
between the simulated and measured electrical Figure 1(b-c) depicts the general structure of our
parameters (i.e., RF index, impedance and loss), but also mmWave-optic modulator, which follows the same
between the electro-optic responses predicted from design as most broadband TFLN modulators
measured electrical parameters and the actually demonstrated to date [26]. The modulator consists of a
measured electro-optic responses. Based on the proposed ground-signal-ground (GSG) copper transmission line
and an optical Mach-Zehnder interferometer (MZI). The The relative modulation efficiency, EO S21, is simply
modulator is based on an x-cut TFLN on top of thermal |m(ω)|2 in dB scale since it relates the power roll-off,
oxide on Si substrate, such that a combination of which could be written as:
transverse-electric (TE) optical modes and in-plane
 − 2 ⋅ S 21 ⋅ cos ( β opt L ) + 1
2 μ
electrodes allows the exploitation of the largest electro- 2 S 21
 (2)
optic tensor component r33 of LN. EO S 21 10 log (1 − H )
=
 ( ) ( ) 
2 2

 ln S 21 + β μ
opt L 
There are two main performance metrics for a mmWave-
optic modulator, which we focus on optimizing in the ω (3)
β opt
= μ
( nm − no )
rest of this paper: (1) the electro-optic S21 (EO S21) c
parameter, which depicts the relative modulation
Z C − Z in (4)
efficiency roll-off with reference to the DC or low- H=
Z C + Z in
frequency (1 GHz in this work) electro-optic response;
and (2) the RF Vπ, which indicates the absolute where S21 is the electrical forward transmission
modulation efficiency at a certain frequency. The 3-dB coefficient in linear scale, which represents the
and 6-dB bandwidths of a modulator refer to the contribution from RF loss; β µopt is the wavevector
frequencies at which EO S21 drops by the respective dB (velocity) mismatch term as defined in Eq. (3); Η
numbers. At 3-dB and 6-dB points, the RF Vπ values are characterizes the mmWave reflection due to mismatch
1.414× and 2× the low-frequency Vπ respectively, since between the input impedance of the transmission line, Zin,
the input electrical power scales quadratically with and that of the driving circuit, ZC, as defined in Eq. (4).
modulation voltage. Therefore both 3-dB and 6-dB
bandwidths are relevant metrics for practical Ideally, when velocity-matching and impedance-
applications. matching conditions are met, both the mismatch terms H
and β µopt L are equal to zero. The EO S21 is then only
The modulation performance at high RF frequencies is determined by the electrical S21 value. Numerically, the
mainly limited by three factors [Fig. 1(b)]. First, the 3-dB and 6-dB electro-optic bandwidths correspond to
phase velocity of mmWave should be matched with the the frequencies at which the electrical S21 responses roll-
optical group velocity, such that the same electrical off by 6.41 dB and 13.8 dB, respectively. The EO S21
signal could be continuously applied to the propagating response rolls off slower than electrical S21 since the
optical pulse resulting in an efficient modulation. Second, former captures the average modulation voltage along
the mmWave loss should be minimal such that the the transmission line. In contrast, the latter corresponds
modulation signal could remain significant throughout to the transmitted electrical power at the output port.
the modulator. Third, the impedance of the transmission
line should be matched with that of external electronics The absolute RF Vπ values could then be calculated from
(typically 50 Ω) to avoid power reflection at the input the EO S21 parameter and the low-frequency Vπ value
port. Quantitatively, the effective modulation voltage using:
(normalized to input voltage) averaged along the V= Vπ, LF × 10− EO S21 20
(5)
transmission line can be expressed as: π, RF

The DC/low-frequency Vπ values in most TFLN

2 Z in ( Z C + Z 0 ) F+ + ( Z C − Z 0 ) F-
m(ω ) = (1) modulators reported to date follow voltage-length
Z in + Z C ( Z C + Z 0 )eγ m L + ( Z C − Z 0 )e-γ m L products (VπL) in the range of 2-3 V∙cm [11, 13, 14], in
order to achieve a good balance between electro-optic
where ω is the mmWave frequency,
overlap and metal-induced optical loss. The VπL product
Z C + Z 0 tanh(γ m L)
Z in = Z 0 is the transmission line input in the devices characterized in this work is 2.3 V∙cm at
Z 0 + Z C tanh(γ m L) DC and 2.67 V∙cm at 1 GHz, consistent with other
impedance, Z0 is the characteristic impedance of the previous demonstrations. The VπL product could
transmission line, ZC is the impedance of the source and potentially be further reduced by material engineering or
termination (usually 50 Ω), novel modulation structures, while the design
ω
± γ m L - j n0 L ω methodologies introduced in this paper for mmWave-
F± =(1 − e c
) / (±γ m L − j no L) corresponds to the band operation shall remain largely the same.
c
ω Following the above theory, the requirements and trade-
forward/backward propagating waves, γ= α m + j nm
m
c offs below should be considered to design a broadband
is the microwave propagation constant (nm is the TFLN mmWave-optic modulator:
mmWave effective index, αm is the loss coefficient), no
(1) Velocity matching between optical and mmWave
is the optical group refractive index, L is the modulation
could be achieved by fine-tuning the buried and cladding
length, c is the speed of light in vacuum.
oxide thickness. A thicker buried (cladding) oxide layer
pushes the mmWave mode [Fig. 1(d)] away from the frequencies below 100 GHz [14, 16, 27, 28], which could
high-index Si substrate (low-index air), therefore potentially be further extended into the upper mmWave
decreasing (increasing) the mmWave effective index. bands using smaller capacitive loading periods to
increase the cut-off frequency. Nevertheless, the design
We numerically simulate the electrical characteristics of
methodology and characterization techniques introduced
the mmWave transmission line using Finite Element
in this paper shall still apply to such more advanced
Methods (FEM, Ansys HFSS) and extract the mmWave
modulator designs.
effective index (nm) from the numerically converged (a) (b)
propagation constant. Figure 2(a) shows the simulated 2.45
Clad 0.6μm (measurement)
44
Measurement
mmWave dispersion curve for different cladding SiO2 2.4 Clad 0.5μm (simulation)
Clad 0.6μm (simulation) 43 Simulation

mmWave refractive index

Clad 0.8μm (simulation)
thicknesses. Our model assumes a dielectric constant of
2.35
no=2.26 42

Impedance (Ω)
2.3
5.5 for the cladding SiO2 from plasma-enhanced 2.25
41

chemical vapor deposition (PECVD), which is inferred 2.2

40

from the actual electrical measurements to be discussed 2.15 39

in Section 3. The optical group index (no) does not 2.1

0 100 200 300
38
0 100 200 300
Frequency (GHz)
change significantly for the fundamental TE mode in our
Frequency (GHz)
(c) (d)

TFLN waveguides, and is simulated to be 2.26 [dashed

15 50 -5.0 16
RF Loss
Impedance
green line in Fig. 2(a)] (Ansys Lumerical Mode) and is
14 45 -5.5 15
EO S21

RF Loss (dB/cm)
insensitive to the buried/cladding oxide thickness. At a
13 -6.0

Impedance (Ω)

RF Loss (dB/cm)
14

EO S21 (dB)
40
12 -6.5
cladding thickness of 600 nm and a buried oxide
13
35
11 -7.0
thickness of 2 µm [solid yellow line in Fig. 2(a)], the
12
30 -7.5
10
mmWave transmission line exhibits phase velocities
11

9 25 -8.0 10
well matched with the optical group velocity throughout 10 20 30 40
Signal width (μm)
50 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
Metal thickness (μm)

the mmWave band. This set of parameters is used for the Fig. 2. (a) Simulated mmWave effective (phase) indices for various
actual devices in this work. cladding thicknesses (solid lines) and the actual indices extracted from
measured s-parameters (circle), as functions of frequency. The green
(2) The signal width of the mmWave transmission line dashed line shows the simulated optical group index, indicating good
should be carefully chosen to balance the trade-off velocity matching in the fabricated device. (b) Simulated and extracted
characteristic impedance of the device. (c) Simulated RF loss and
between impedance matching and RF loss. According to characteristic impedance, as well as calculated EO S21 (assuming
common transmission line theory, the characteristic velocity matching) for various signal widths at 250 GHz. (d) Simulated
impedance Z 0 = ( L / C )1/ 2 is determined by the capacitance RF loss versus metal thickness at 250 GHz.
per unit length C and the inductance per unit length L. (3) A thicker metal layer could be used to reduce the
Using a wider signal line could effectively reduce the Ohmic loss of the transmission line, as far as the
Ohmic loss, but at the same time increases the fabrication process allows. However, metal thicknesses
capacitance between the signal and ground and reduces beyond 800 nm do not quite benefit operations at upper
the inductance, leading to lower and often unmatched mmWave frequencies since the skin depth at, for
impedance. example 250 GHz, is only 150 nm, as Fig. 2(d) shows.
We use the same HFSS model to extract the RF loss and In our devices to be discussed next, the metal thickness
characteristic impedance at 250 GHz as a function of the is chosen to be 800 nm. Metal thickness also slightly
signal width, which clearly shows a trade-off between shifts the mmWave velocity and impedance, which could
the two [Fig. 2(c)]. Due to meshing-related uncertainties be fine-tuned using the methods discussed in the above
of the simulation software, the RF loss values fluctuate points 1-2.
slightly, which does not affect the overall trend. It should (4) A longer modulation length leads to faster EO S21
be noted that impedance mismatch within a certain range roll-off due to larger RF losses, thus smaller 3-dB and 6-
does not significantly affect the modulation performance. dB bandwidths. However, for practical applications at
For example, a 40-Ω characteristic impedance certain mmWave frequencies, the more important
corresponds to ~ 11% drop in the modulation voltage. As parameter is the absolute modulation efficiency, i.e., RF
a result, the best electro-optic response may not occur at Vπ. In the ideal case where velocity is perfectly matched,
the exact impedance-matched point. The green line of a longer modulation length always leads to lower RF Vπ,
Fig. 2(c) shows our calculated electro-optic responses at since even the much-attenuated sections still contribute
250 GHz based on the calculated RF loss and impedance to the total modulation effect. However, longer
(assuming velocity matching), indicating a maximum electrodes in an actual device are more sensitive to any
electro-optic response at a signal width of 20 µm, which velocity mismatches, as can be seen in Eq. (2). In this
we use in the actual devices. More advanced paper, we evaluate modulators with two different lengths,
transmission line structures like capacitive loading have i.e., 5.8 mm and 10.8 mm, which show a good balance
recently shown the capability to break this trade-off at
between the achievable RF Vπ and the tolerance to as shown in Fig. 3(b). The electrical loss comes from two
fabrication variations. origins: the conductor loss αc, which is typically
proportional to the square root of frequency, and the
We note that actual electrical parameters often deviate
dielectric loss αd, which goes linearly with frequency.
from the simulated values due to deviations in actual
The total electrical loss for an electrode length L could
fabrication parameters and material properties.
then be modelled as (α c f + α d f ) L + A , where A is the
Therefore, it is important to perform a careful electrical
evaluation of the fabricated mmWave transmission lines intercept loss at DC originated from impedance
and fine-tune the theoretical model to achieve the desired mismatch and is equal to 2.6 dB in this case. Based on
modulation performances in the full electro-optic our measured results, we estimate that αc=0.042 dB cm-1
devices, which we discuss next. GHz-1/2 and αd=0.025 dB cm-1 GHz-1. The main loss
mechanism at frequencies below 50 GHz is conductor
3. Experiments
loss, consistent with previous reports [14]. However, at
A. Device fabrication upper mmWave bands, the dielectric loss contribution
becomes more significant and should not be ignored, as
The devices are fabricated on a TFLN wafer from
the gap between the two fitted lines in Fig. 3(b) shows.
NANOLN consisting of a 600 nm thick x-cut LN bonded
The measured electrical loss is even larger than our
on top of thermal oxide (2 μm thick) on a 500-μm-thick
model at above 200 GHz, possibly because the dielectric
silicon substrate. We use electron-beam lithography
loss tangent increases at higher frequencies while our
(EBL) to define the optical waveguide patterns in
model assumes a fixed loss coefficient. At these
Hydrogen Silsesquioxane (HSQ) and transfer the
frequencies there also exist resonance-like features
patterns into TFLN by dry etching 300 nm of the LN film
likely due to reflections at the two ends of the
using argon ion (Ar+)-based reactive ion etching (RIE)
transmission lines. The extracted αd gives rise to an
process [13]. Then, we deposit a metal layer (800-nm-
effective dielectric loss tangent of 0.025 for this
thick copper terminated with 30-nm-thick gold on top)
transmission line mode, which is consistent with the loss
to form the transmission lines through aligned
tangents of the dielectrics involved here, i.e., 0.025 for
photolithography, evaporation and lift-off processes.
silicon [29], 0.008 for lithium niobate [30], and 0.000042
The transmission line signal width is 20 μm and the gap
for SiO2 [31].
between electrodes is 5 μm. The rib optical waveguide is
1.2 μm wide on top in the modulation region and 0.8 μm Apart from the RF losses, it is important to extract the
wide for routing, with a 300 nm thick slab. The chip is actual mmWave velocity and impedance in the
cladded with 0.6 μm thick silica by PECVD. The silica fabricated devices, in order to fulfill and verify the
cladding is selectively removed at probe contact areas by design guidelines in Section 2. The mmWave index nm
another photolithography process followed by RIE is extracted by nm =
βc
(βis the propagation constant
etching. Finally, the edge of the chip is diced and 2πω
polished to enhance the coupling between fiber and chip. extracted from the measured phase response). The
Figure 1(e) shows the top view of a fabricated mmWave- (1 + S11 ) 2 − S 212
optic modulator under an optical microscope. impedance is extracted by Z in = Z C . The
(1 − S11 ) 2 − S 212
B. Characterizations of device electrical properties. circles in Fig. 2(a-b) shows the extracted nm and Zin
We first perform a detailed electrical analysis of the values at frequencies up to 300 GHz, both of which
fabricated mmWave-optic modulators. We measure the match well with the designed numbers. The difference
reflection (S11) and transmission (S21) s-parameters of the between the measured and target nm is 0.02 at 150 GHz
transmission line from 10 MHz to 325 GHz using an and 0.04 at 250 GHz, allowing for a reasonably well-
Agilent vector network analyzer (VNA) and frequency matched velocity within the entire frequency range of
extension modules in the 10 MHz-69 GHz, 65-110 GHz, interest. The extracted Zin is 41.5 Ω at 250 GHz, which
90-140 GHz and 220-325 GHz bands, respectively. The is also close to the simulated result (40.5 Ω). The
VNA is calibrated using short-open-load-thru (SOLT) extracted values see relatively large fluctuations at
standards. A pair of mmWave GSG probes is used to certain frequencies since the extension modules exhibit
launch mmWave signals into the input port of the degraded performances near the edges of each
transmission line and collect them from the output port. measurement band. The remaining differences between
Figure 3(a) shows the measured S11 and S21 parameters the measured and simulated nm and Zin values could
of the 5.8-mm and 10.8-mm devices, respectively. As result from variations in the fabricated device’s
expected, the transmitted power rolls off at higher structural parameters and measurement uncertainties.
frequencies, with a measured loss of 1.3 dB/mm at 250 Nevertheless, the ability to extract the actual velocity,
GHz. From the measured s-parameters, we can extract impedance and loss information of the mmWave
the electrical loss coefficient α of the transmission line, transmission line serves as an invaluable asset for
predicting and explaining the modulation performances (a)

at mmWave frequencies, as we show next. RF Multiplier

RF Generator
(a) (b) 50 Ω PD
0 0 45
S11 S21 5.8mm Measured
-5 40
S21 10.8mm -5 Fit (both conductor and dielectric loss )
Laser OSA
-10 35 Fit (only conductor loss)
Polarization
Controller LN Chip under test

RF Loss (dB/cm)
-10 30
-15
EE S21(dB)

EE S11(dB)
-1 -1/2
25 αc = 0.42 dBcm GHz
-20 -15
20 αd = 0.025 dBcm GHz
-1 -1
(b)
-25
-20 15
-30
10
-25
-35 5
-40 -30 0
0 100 200 300 0 100 200 300
Frequency (GHz) Frequency (GHz)

Fig.3. (a) Measured s-parameters of mmWave transmission lines with

lengths of 5.8 mm and 10.8 mm. (b) Extracted and fitted electrical
loss coefficient of the transmission line as a function of frequency.

C. Characterizations of device electro-optic properties.

Fig. 4. Schematic diagram (a) and photo (b) of the measurement setup
We measure the electro-optic responses of our TFLN for characterizing electro-optic responses at frequencies up to 325 GHz.
mmWave-optic modulators using the setup shown in Fig.
4. A wavelength-tunable laser source (Santec TSL-550, We could therefore calculate the RF Vπ using Eq. (7) [15]:
1500-1630 nm) is used to input light in the telecom L- 1
band. A 3-paddle fiber polarization controller (FPC) is Vπ, RF = πV p Ps (7)
4
used to ensure TE mode excitation. Light is coupled into
and out from the chip using a pair of lensed fibers. where Vp is the peak voltage of the input mmWave signal,
Measurements are separately conducted in four which is measured using a mmWave power meter at each
frequency bands. For the lowest band (< 67 GHz), frequency point and carefully calibrated by excluding the
modulation electrical signals are directly generated from losses from the input mmWave waveguide and the probe.
a RF generator (MG3697C, Anritsu). For higher
Based on the extracted RF Vπ values and the Vπ at 1 GHz
frequencies (> 67 GHz), microwave signals are
(Vπ, LF), we could further calculate the EO S21 using Eq.
generated by up-conversion, where 250 kHz – 20 GHz
(5).
signals (Agilent E8267D) are up-converted and
amplified by frequency multipliers in the respective Figure 5 shows the extracted RF Vπ (a-b) and EO S21 (c-
bands, i.e., 65 – 110 GHz, 90 – 140 GHz and 220 – 325 d) at frequencies up to 325 GHz for the 10.8-mm and the
GHz. A pair of mmWave probes are used to deliver the 5.8-mm device (blue dots), plotted together with the ones
modulation signal to the input port of the transmission calculated from the measured electrical parameters (red
line, and to terminate the output port with a 50-Ω load. lines) in Section 3B following Eq. (2). The directly
The modulator is biased at the quadrature point, resulting measured electro-optic responses show more
in an output optical signal with two sidebands separated fluctuations mostly due to uncertainties in the quoted Vp
from the carrier by the mmWave frequency [inset of Fig. in Eq. (7). The mmWave multiplexer often generates
4(a)]. The electro-optic response is tested by monitoring spurious harmonics away from the intended frequency,
the power ratio between the sideband and the carrier, which may also be counted by the power meter.
using an optical spectrum analyzer (OSA), which we Nonetheless, the direct electro-optic measurements show
define as the normalized sideband power Ps: excellent agreement with the calculated results, thanks to
the precise measurements of velocity, impedance and
Psideband
Ps = (6) mmWave loss we have performed.
Pcarrier
In Fig. 4(a), we could see that the upper and lower Our measurement results indicate that the 5.8-mm device
sidebands show slightly different powers. This is features ultrahigh 3-dB and 6-dB electro-optic
because at ultrahigh modulation frequencies, the bandwidths of 170 GHz and 295 GHz, respectively [Fig.
modulated sidebands are significantly away from the 5(d)]. The bandwidths of the 10.8-mm device are
carrier such that they experience a shifted bias point in relatively lower, i.e.,100 GHz (3 dB) and 175 GHz (6
the unbalanced MZI. In this case, we take the average dB), due to more severe RF attenuation [Fig. 5(c)]. Still,
power of the two sidebands, which allows the below the 10.8-mm device provides lower RF Vπ than that of
analysis to hold in presence of the unbalanced sidebands. the 5.8-mm device throughout the entire mmWave band
for the reasons discussed in Section 2 and could be more
appealing in most application scenarios. The measured
RF Vπ values are 7.3 V for the 10.8-mm device and 8 V
for the 5.8-mm device [Fig. 5(a-b)] at an ultrahigh
mmWave frequency of 250 GHz, making these devices
 highly relevant for practical applications. In comparison, 20
the RF Vπ of a typical commercial LN modulator could 18
Measured

be as high as 15 V at 100 GHz (3.8 V in our device), as 16

14
∆n=0.1, Z0=41.5

measured in [32]. 12 ∆n=0.07, Z0=41.5

RF Vπ (v)
10
(a) (b) 8 ∆n=0.04, Z0=41.5
16 16
Length 10.8mm Length 5.8mm
14 14 EO measurement 6
EO measurement ∆n=0, Z0=41.5
12 calculated values 12
calculated values 4
∆n=0, Z0=50
2
10 10
RF Vπ (V)

RF Vπ (V)
0
8 8 0 1 2 3 4 5
Device Length (cm)
6 6

4 4
Fig. 6. Simulated RF Vπ vs. device length for various velocity- and
impedance-mismatch conditions at 250 GHz. The yellow dots
2 2

correspond to the 5.8-mm and 10.8-mm devices in this work.

0 0
0 100 200 300 0 100 200 300
Frequency (GHz) Frequency (GHz)
(c) (d)
0 0 5. Conclusions
-2 3dB -2
-4
3dB
In this paper, we report a detailed theoretical and
experimental analysis of the mmWave-optic modulation
6dB -4
-6
EO S21(dB)
EO S21 (dB)

6dB
-8

-10
-6
performances of TFLN modulators at frequencies up to
-12
Length 10.8mm
-8
325 GHz. We show that the ability to reliably extract the
actual mmWave velocity and impedance in the
Length 5.8mm
-14 Measurement -10 measurement
Calculated calculated
-16
0 100 200 300
-12
0 100 200 300 fabricated devices, is key to an accurate and systematic
design optimization at these ultrahigh frequencies. We
Frequency (GHz) Frequency(GHz)

Fig. 5. (a)-(b) Measured and calculated modulator RF Vπ of the 10.8 show that an optimized 5.8 mm long modulator could
mm (a) and 5.8 mm (b) devices. (c)-(d) Calculated and measured provide a measured 3-dB electro-optic bandwidth of 170
electro-optic responses of the 10.8 mm (c) and 5.8 mm (d) devices. The
measured lines are extracted directly from the electro-optic sideband GHz and a 6-dB bandwidth of 295 GHz. The devices
measurements. The calculated lines are based on the nm and Zin values also show RF half-wave voltages as low as 6.3 V at an
from the measured electrical s-parameters. ultrahigh modulation frequency of 250 GHz. The design
4. Discussions and characterization methodologies in this work could be
readily applied to more advanced modulator
The absolute RF Vπ is usually more important for architectures as well as other material platforms. The
practical applications as it determines the actual ultra-broadband low Vπ mmWave-optic modulators
mmWave-optic conversion efficiency. As presented could become key elements in future mmWave systems
above, a longer modulator length should always lead to for telecommunication, short-range radar and sensors
lower Vπ as long as the velocity matching condition applications.
could be fulfilled, which, however, becomes Funding. National Natural Science Foundation of China (61922092);
increasingly more vulnerable for longer electrodes and Research Grants Council, University Grants Committee (CityU
higher frequencies. Figure 6 shows the calculated RF Vπ 11204820, CityU 21208219, T42-103/16-N); Croucher Foundation
values as functions of device length and velocity (9509005); City University of Hong Kong (9610402, 9610455);
Central Research Fund (CRF), Agency for Science, Technology and
(index)/impedance mismatch at 250 GHz. The current
Research (A*STAR); Harvard Quantum Initiative (HQI) Postdoctoral
devices (yellow dots) operate along the yellow line with Fellowship
a slight index mismatch of Δn = 0.04. Along this line, the
Disclosures. The authors declare no conflicts of interest.
RF Vπ first decreases and then increases as the modulator
length increases, indicating that there exists an optimal Data availability. The data in this study are available from the
modulation length that provides the lowest RF Vπ (in this corresponding authors on reasonable request in the paper.
case, 6.27 V at 1.15 cm). If perfect velocity matching
could be achieved, the RF Vπ shall monotonically References
decrease following the red line, potentially achieving an 1. J. M. Lien, N. Gillian, G. M. E. Karagozler, P. Amihood, C.
RF Vπ of 4.17 V at 250 GHz for a device length of 1.5 Schwesig, E. Olson, and H. Raja, "Soli: Ubiquitous Gesture Sensing
cm. This performance envelope could be further with Millimeter Wave Radar," Acm Trans. Graph. 35, pp. 1-19 (2016).
improved by satisfying the velocity- and impedance- 2. D. Marpaung, J. P. Yao, and J. Capmany, "Integrated microwave
photonics," Nat. Photonics 13, 80-90 (2019).
matching conditions simultaneously (blue line), which is 3. H. Murata, "Millimeter-Wave-Band Electro-Optic Modulators
currently not possible for a reasonably wide signal line, Using Antenna-Coupled Electrodes for Microwave Photonic
but could potentially be achieved using a capacitive- Applications," J. Lightwave Technol. 38, 5485-5491 (2020).
loaded electrode structure with small enough periods to 4. T. Nagatsuma, S. Hisatake, M. Fujita, H. H. N. Pham, K. Tsuruda,
S. Kuwano, and J. Terada, "Millimeter-Wave and Terahertz-Wave
extend the cutoff frequencies into upper mmWave Applications Enabled by Photonics," IEEE J. Quantum Elect. 52
frequencies [14, 16, 27, 28]. (2016).
5. Y. Ogiso, J. Ozaki, Y. Ueda, H. Wakita, M. Nagatani, H. Yamazaki, narrowband Bragg grating filter on thin-film lithium niobate," Opt. Lett.
M. Nakamura, T. Kobayashi, S. Kanazawa, Y. Hashizume, H. Tanobe, 46, 2730-2733 (2021).
N. Nunoya, M. Ida, Y. Miyamoto, and M. Ishikawa, "80-GHz 23. J. X. Zhou, R. H. Gao, J. T. Lin, M. Wang, W. Chu, W. B. Li, D.
Bandwidth and 1.5-V Vπ InP-Based IQ Modulator," J. Lightwave F. Yin, L. Deng, Z. W. Fang, J. H. Zhang, R. B. Wu, and Y. Cheng,
Technol. 38, 249-255 (2020). "Electro-Optically Switchable Optical True Delay Lines of Meter-
6. C. Han, M. Jin, Y. Tao, B. Shen, H. Shu, and X. Wang, "Ultra- Scale Lengths Fabricated on Lithium Niobate on Insulator Using
compact silicon modulator with 110 GHz bandwidth," in Optical Fiber Photolithography Assisted Chemo-Mechanical Etching," Chin. Phys.
Communications Conference (2022), pp. 1-3. Lett. 37, 8 (2020).
7. S. Ummethala, T. Harter, K. Koehnle, Z. Li, S. Muehlbrandt, Y. 24. C. Op de Beeck, F. M. Mayor, S. Cuyvers, S. Poelman, J. F.
Kutuvantavida, J. Kemal, P. Marin-Palomo, J. Schaefer, A. Tessmann, Herrmann, O. Atalar, T. P. McKenna, B. Haq, W. Jiang, J. D. Witmer,
S. K. Garlapati, A. Bacher, L. Hahn, M. Walther, T. Zwick, S. Randel, G. Roelkens, A. H. Safavi-Naeini, R. Van Laer, and B. Kuyken, "III/V-
W. Freude, and C. Koos, "THz-to-optical conversion in wireless on-lithium niobate amplifiers and lasers," Optica 8, 1288-1289 (2021).
communications using an ultra-broadband plasmonic modulator," Nat. 25. Z. X. Chen, Q. Xu, K. Zhang, W. H. Wong, D. L. Zhang, E. Y. B.
Photonics 13, 519-524 (2019). Pun, and C. Wang, "Efficient erbium-doped thin-film lithium niobate
8. E. Heidari, H. Dalir, F. M. Koushyar, B. M. Nouri, C. Patil, M. waveguide amplifiers," Opt. Lett. 46, 1161-1164 (2021).
Miscuglio, D. Akinwande, and V. J. Sorger, "Integrated ultra-high- 26. C. Wang, M. Zhang, B. Stern, M. Lipson, and M. Loncar,
performance graphene optical modulator," arXiv:2109.07476 (2021). "Nanophotonic lithium niobate electro-optic modulators," Opt.
9. D. Zhu, L. B. Shao, M. J. Yu, R. Cheng, B. Desiatov, C. J. Xin, Y. Express 26, 1547-1555 (2018).
W. Hu, J. Holzgrafe, S. Ghosh, A. Shams-Ansari, E. Puma, N. Sinclair, 27. X. Liu, B. Xiong, C. Sun, Z. Hao, L. Wang, J. Wang, Y. Han, H.
C. Reimer, M. A. Zhang, and M. Loncar, "Integrated photonics on thin- Li, and Y. Luo, "Capacitively-loaded Thin-film Lithium Niobate
film lithium niobate," Adv. Opt. Photonics 13, 242-352 (2021). Modulator with Ultra-flat Frequency Response," IEEE Photonic
10. A. R. Ahmed, S. Nelan, S. Y. Shi, P. Yao, A. Mercante, and D. W. Technol. Lett., 1-1 (2022).
Prather, "Subvolt electro-optical modulator on thin -film lithium 28. X. Liu, B. Xiong, C. Sun, J. Wang, Z. Hao, L. Wang, Y. Han, H.
niobate and silicon nitride hybrid platform," Opt. Lett. 45, 1112-1115 Li, and Y. Luo, "Sub-terahertz bandwidth capactively-loaded thin-film
(2020). lithium niobate electro-optic modulators based on an undercut
11. X. C. Liu, B. Xiong, C. Z. Sun, J. Wang, Z. B. Hao, L. Wang, Y. J. structure," Opt. Express 29, 41798-41807 (2021).
Han, H. T. Li, J. D. Yu, and Y. Luo, "Wideband thin-film lithium 29. M. Wan, Y. Yashchyshyn, X. Li, K. Godziszewski, D.-W. Sun, and
niobate modulator with low half-wave-voltage length product," Chin. J. Sheridan, "Characterization of silicon in the terahertz, " SPIE (2020),
Opt. Lett. 19, 6 (2021). 11279.
12. M. Y. Xu, M. B. He, H. G. Zhang, J. Jian, Y. Pan, X. Y. Liu, L. F. 30. M. Lee, "Dielectric constant and loss tangent in LiNbO3 crystals
Chen, X. Y. Meng, H. Chen, Z. H. Li, X. Xiao, S. H. Yu, S. Y. Yu, and from 90 to 147 GHz," Appl. Phys. Lett. 79, 1342-1344 (2001).
X. L. Cai, "High-performance coherent optical modulators based on 31. N. Chudpooti, N. Duangrit, A. D. Burnett, J. R. Freeman, T. B. Gill,
thin-film lithium niobate platform," Nat. Commun. 11, 3911 (2020). C. Phongcharoenpanich, U. Imberg, D. Torrungrueng, P.
13. C. Wang, M. Zhang, X. Chen, M. Bertrand, A. Shams-Ansari, S. Akkaraekthalin, I. D. Robertson, and N. Somjit, "Wideband dielectric
Chandrasekhar, P. Winzer, and M. Loncar, "Integrated lithium niobate properties of silicon and glass substrates for terahertz integrated
electro-optic modulators operating at CMOS-compatible voltages," circuits and microsystems," Mater. Res. Express 8 (2021).
Nature 562, 101-104 (2018). 32. P. T. Dat, Y. Yamaguchi, K. Inagaki, M. Motoya, S. Oikawa, J.
14. P. Kharel, C. Reimer, K. Luke, L. He, and M. Zhang, "Breaking Ichikawa, A. Kanno, N. Yamamoto, and T. Kawanishi, "Transparent
voltage–bandwidth limits in integrated lithium niobate modulators Fiber–Radio–Fiber Bridge at 101 GHz using Optical Modulator and
using micro-structured electrodes," Optica 8, 357-363 (2021). Direct Photonic Down-Conversion," in Optical Fiber Communications
15. A. J. Mercante, S. Y. Shi, P. Yao, L. L. Xie, R. M. Weikle, and D. Conference (2021), pp. 1-3.
W. Prather, "Thin film lithium niobate electro-optic modulator with
terahertz operating bandwidth," Opt. Express 26, 14810-14816 (2018).
16. G. X. Chen, K. X. Chen, R. F. Gan, Z. L. Ruan, Z. Wang, P. C.
Huang, C. Lu, A. P. T. Lau, D. X. Dai, C. J. Guo, and L. Liu, "High
performance thin-film lithium niobate modulator on a silicon substrate
using periodic capacitively loaded traveling-wave electrode," APL
Photonics 7, 026103 (2022).
17. P. O. Weigel, J. Zhao, K. Fang, H. Al-Rubaye, D. Trotter, D. Hood,
J. Mudrick, C. Dallo, A. T. Pomerene, A. L. Starbuck, C. T. DeRose,
A. L. Lentine, G. Rebeiz, and S. Mookherjea, "Bonded thin film
lithium niobate modulator on a silicon photonics platform exceeding
100 GHz 3-dB electrical modulation bandwidth," Opt. Express 26,
23728-23739 (2018).
18. H. Feng, K. Zhang, W. Sun, Y. Ren, Y. Zhang, W. Zhang, and C.
Wang, "Ultra-high-linearity integrated lithium niobate electro-optic
modulators," arXiv:2202.12739 (2022).
19. Y. He, Q. F. Yang, J. W. Ling, R. Luo, H. X. Liang, M. X. Li, B.
Q. Shen, H. M. Wang, K. Vahala, and Q. Lin, "Self-starting bi-
chromatic LiNbO3 soliton microcomb," Optica 6, 1138-1144 (2019).
20. M. Zhang, B. Buscaino, C. Wang, A. Shams-Ansari, C. Reimer, R.
R. Zhu, J. M. Kahn, and M. Loncar, "Broadband electro-optic
frequency comb generation in a lithium niobate microring resonator,"
Nature 568, 373-377 (2019).
21. A. Prencipe, M. A. Baghban, and K. Gallo, "Tunable
Ultranarrowband Grating Filters in Thin-Film Lithium Niobate," ACS
Photonics 8, 2923-2930 (2021).
22. K. Abdelsalam, E. Ordouie, M. G. Vazimali, F. A. Juneghani, P.
Kumar, G. S. Kanter, and S. Fathpour, "Tunable dual-channel ultra-