Antenna-coupled integrated millimeterwave modulators and resonant

electro-optic frequency combs

Aleksei Gaier1, 2 , Karen Mamian1 , Shima Rajabali3, 4 , Yazan Lampert1, 2 , Jiawen Liu1, 2 , Leticia
Magalhaes3 , Amirhassan Shams-Ansari5 , Marko Lončar3 , and Ileana-Cristina Benea-Chelmus1,2
1
Hybrid Photonics Laboratory, École Polytechnique Fédérale de Lausanne (EPFL), CH-1015, Switzerland
2
Center for Quantum Science and Engineering (QSE), CH-1015, Switzerland
arXiv:2505.04585v1 [physics.optics] 7 May 2025

3
Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA
4
Department of Quantum and Computer Engineering, Delft University of Technology, Netherlands
5
DRS Daylight Solutions, 16465 Via Esprillo, CA, USA

May 8, 2025

Abstract

The rapid growth of global data traffic is accelerating the need for ultra-broadband communication technologies,
particularly in inter-data center links, cloud infrastructure, and emerging 6G wireless systems. Applications such as
optical computing and quantum information processing further demand fast, scalable ways to interface optical and
electronic signals. Integrated electro-optic modulators offer a compact and efficient solution, but extending their
operation into the millimeterwave (mmWave) and terahertz regimes while providing wide operational bandwidths
and compatibility with wireless mmWave signals is difficult. Key challenges include high transmission line losses
and bulky electrical packaging, limiting the performance and scalability of these systems.
Here, we demonstrate a wireless and wideband electro-optic modulation architecture that directly interfaces
wireless mmWave signals with optical ones, eliminating the need for impedance-matched mmWave probes and
cables. By integrating an on-chip antenna directly with a co-designed traveling-wave transmission line on thin-film
lithium niobate, we achieve spectrally flat phase modulation across the WR9.0 (82–125 GHz) and WR2.8 (240–380
GHz) bands. The wideband nature of our modulator enables the device to function as a high-speed mmWave
detector. By modulating the mmWave carrier at frequencies up to 6 GHz, we demonstrate the device’s flat and
wide detection bandwidth—critical for 6G communication and high-speed mmWave sensing. Beyond traveling-
wave detection, our wireless platform enables resonant electro-optic frequency comb generation with mode spacing
of 123.2 GHz and 307.9 GHz. Extracted single-photon electro-optic coupling rates of geo = 2π × 10.61 kHz and
25.99 kHz, at 123.2 and 307.9 GHz, respectively, demonstrate favorable scaling with mmWave frequency. These
results establish a new class of broadband, wireless electro-optic devices for high-speed modulation, detection, and
frequency comb generation, with promising applications in communications, sensing, and quantum technologies.

1 Introduction
The rapid expansion of cloud services, AI workloads, and global data exchange has led to an unprecedented surge in
data traffic, placing stringent demands on the underlying communication infrastructure [1]. Exploiting under-utilized

1
frequency bands such as the mmWave (30-300 GHz) and the terahertz (300 GHz - 10 THz) offers a promising path to
meet the demand for this ever-increasing growth [2]. These bands provide access to much larger available bandwidths
than traditional microwave frequencies while offering improved robustness to atmospheric turbulence compared to
free-space optical communications at telecom frequencies. However, all-electronic transmitters, receivers and mixers
needed to establish a mmWave/terahertz communication link, typically based on high-frequency electronic devices
such as high-electron-mobility transistors [3] or Schottky barrier diodes [4], face fundamental challenges. In particular,
the signal integrity and the bandwidth of these systems are limited due to technical issues. For example, a typical way
to generate high-frequency signals is to implement the frequency extenders [4]. However, these components generate
several harmonics of the fundamental frequency, and the insufficient suppression of undesired harmonics compromises
the purity of the generated signal [5]. Similarly, the bandwidth of the intermediate frequencies (IF) of commercial
all-electronic mixers is about a few tens of gigahertz, limiting the data modulation (demodulation) speed at the sender
(receiver) end [6]. Moreover, high power consumption for signal generation and amplification, and susceptibility to
electrostatic discharge (ESD) impose further constraints on all-electronic systems [7].
To address these limitations while achieving large operating bandwidth, photonic technologies have emerged as
a powerful alternative [8, 9]. By enabling the interaction between microwave signals and optical carriers, photonics
offers access to significantly broader bandwidths, reduced long-distance losses, and multiplexing capabilities. This
field, often referred to as radio-frequency (RF) photonics, has already demonstrated transformative potential in high-
speed communications [10] and advanced sensing systems [11]. A key component in current communication networks
that overcomes these problems at the receiver end is an integrated electro-optic modulator, which converts microwave
signals to the optical domain within a compact and power-efficient chip. Of particular interest for next generation
communications systems are wireless modulators which are designed to directly capture free-space signals and encode
them onto optical carriers [2]. By providing a direct interface between the wireless signals used in communications
systems such as 5G, with optical data transmission infrastructure, these modulators enable high-capacity links between
wireless access points and long-haul fiber networks.
To further meet the demands of next-generation technologies such as the emerging 6G networks, there is a grow-
ing interest in extending the operation of electro-optic modulators from traditional microwave frequencies into the
millimeter-wave (mmWave) and terahertz (THz) regimes [2]. Among various mechanisms explored for phase and
intensity modulation, second-order (χ(2) ) nonlinearities stand out due to their intrinsically fast electronic response,
enabling modulation bandwidths that far exceed the limitations of carrier-based modulators. High-speed modulation
in mmWave and THz spectral bands would not only benefit optical communication but also enable critical advances
in quantum optics [12, 13, 14], quantum transduction between optical and microwave domains [15], mmWave radar
and ranging systems [16, 11], and quantum computing architectures [17, 18, 19].
Achieving efficient light modulation at mmWave and THz frequencies, while maintaining compact, integrated,
and low-loss device architectures, remains a key technological challenge. This has driven extensive research across
various material platforms [20]. Material platforms under investigation include lithium niobate [21, 22, 23, 10], hybrid
silicon nitride–lithium niobate [24, 25], hybrid silicon-organic modulators [26, 27, 28, 9], aluminum nitride [29], barium
titanate [30] and lithium tantalate [8].
When it comes to the platform selection for mm-Wave and THz, two properties must be considered, namely the
THz loss, and the broadband nonlinearity of the material. Among these, in terms of THz loss aluminum nitride stands
out due the high frequency of its optical phonon [31]. However, its suffers from relatively low electro-optic coefficients
(r33 = −0.59 pm/V) [29]. In contrast, barium titanate has extremely high Pockels coefficients (r42 = 500 pm/V). How-

2
ever in addition to large THz loss [32], its nonlinearity rolls-off rapidly at modulation frequencies beyond 80 GHz [30].
Organic materials have low terahertz loss and achieve the record-breaking modulation bandwidths [9]. However, it
is important to note that organics suffer from relatively high optical absorption and do not yet form a stand-alone
photonic platform, necessitating integration e.g. with silicon-on-insulator [33, 34]. Thin film lithium niobate (TFLN)
offers both low optical loss, moderate nonlinearities and high-power stability. Recently, TFLN has been used for
both broadband terahertz generation [35, 36], and terahertz detection and beam profiling at 500 GHz [37] as early
demonstrations of hybrid THz-photonics devices.
A critical but relatively underexplored challenge in developing of mmWave electro-optic modulators is efficiently
delivering the mmWave signal on to the chip. Currently, most demonstrations rely on probe-based coupling methods,
which become increasingly fragile, expensive, and difficult to scale at frequencies above a few tens of gigahertz.
These methods experience significant attenuation of high-frequency signals due to decreasing skin depth (scaling
√
with ∝ fmmW [38, 39]). Both the electrical circuit delivering the mmWave signals to the photonic chip and RF
circuit on the chip contribute to this loss. Another complication arises from increased radiative losses which scales as
∝ w2 fmmW
3
[39, 40], with w being the distance between ground and signal electrodes. This scaling leads to elevated
cross-talk at higher frequencies. Additionally, impedance matching between the coaxial cables, the RF probes, and
the on-chip transmission line (typically 50 Ω) presents a significant challenge in mmWave and THz regimes. We would
note that the current mmWave delivery technology relies on rectangular waveguides to mitigate radiative and metallic
loss. However, these waveguides are bulky and lack the necessary flexibility for practical application and integration.
These challenges highlight the need for developing practical and low-loss techniques to enhance the efficiency
of electro-optic modulation across the entire signal path. Transitioning from wired to wireless approaches offers a
promising solution by transmitting millimeterwave signals through free space, thereby avoiding ohmic losses, and
directly coupling them into the chip via large aperture, eliminating the need for complex interfaces. Previous demon-
strations [26, 11, 41, 42] have been restricted to a relatively narrow operational bandwidth.
In this work, we propose a wireless and wideband mmWave and terahertz electro-optic modulator on thin film
lithium niobate platform that operates across the WR9.0 (82-125 GHz) and WR2.8 (240-380 GHz) bands (Fig. 1 a).
Our device implements an on-chip antenna coupling free-space terahertz radiation to an on-chip transmission line
eliminating the need for complex and fragile terahertz signal delivery components. To ensure the broadband operation
of this device, the index of the mm-Wave traveling across the transmission line is engineered to match the group index
of the optical signal enabling velocity matching over a wide frequency range. Furthermore, we integrate a hyper-
hemispherical silicon lens on the backside of our TFLN chip (aligned with the on-chip antenna) to improve the coupling
efficiency of the incoming mmWave/terahertz signal (Fig. 1 b). Upon being illuminated by the external mmWave
source, the antenna becomes a mmWave signal generator with a voltage amplitude Va and characteristic impedance
Za that can be tailored through design. This flexibility significantly simplifies the co-design of the transmission line
to minimize optical and mmWave losses while maximizing the modulation efficiency. We note that such miniaturized
mmWave devices are incompatible with conventional characterization techniques, necessitating innovative approaches
to extract the mmWave transmission line parameters. To address this we develop a photonics-based characterization
technique showing excellent agreement between theory and experiment. Our method is a step towards verification of
thin films properties against their bulk counterparts at mmWave frequencies.
Leveraging our flat frequency response we realize a miniaturized high-speed terahertz detector (Fig. 1 c). In
contrast to the state-of-the-art terahertz detectors that rely on pyroelectric effect limiting their speed to a few kHz,
our device operates 6-orders of magnitude faster, reaching several gigahertz range. This dramatic improvement

3
 a b
On-chip antenna
Chip

Input light Antenna-coupled transmission line Modulated output

Free-space
mmWave signal Free-space
Integrated photonic chip mmWave signal

c High-speed mmWave/Terahertz sensing d Electro-optic comb generation

Laser source Specral filter PD

Laser source

Fig. 1: Wideband wireless electro-optic (EO) modulator in TFLN and its applications in high-speed
sensing and frequency comb formation. (a) Conceptual sketch of a wireless electro-optic (EO) modulator, where
a free-space mmWave signal is captured by an on-chip antenna and coupled to a transmission line. This transmission
line runs adjacent to an optical waveguide fabricated in thin-film lithium niobate (TFLN), enabling phase modulation
of the guided light via the electro-optic effect. (b) Illustration of the wireless signal delivery scheme, where an incoming
mmWave or terahertz beam is focused onto an on-chip antenna using a hyper-hemispherical silicon lens to maximize
coupling efficiency. (c) Schematic of a high-speed free-space radiation sensor based on the wireless EO modulator.
A laser pumps the optical waveguide, and incident mmWave or terahertz radiation induces phase modulation in the
optical path, generating an optical sideband in the frequency domain. Then the carrier frequency is filtered, and
then the sideband is detected by a photodiode (PD). (d) Diagram of an EO comb generator architecture utilizing
the wireless EO modulator, where the incoming wireless signal induces a modulated optical spectrum suitable for
applications in spectroscopy and communication.

stems from its operation based on the ultra-fast electro-optic effect with its detection speed currently limited by the
modulation bandwidth of the incoming signal. Finally, we leverage the low-loss nature of TFLN platform and place our
phase modulator in a racetrack cavity (Fig. 1 d). This enhances the effective path length over which the optical signal
interacts with the mmWave, and results in a cascaded sideband generation. Using this, we realize a proof-of-principle
electro-optic frequency comb source with adjustable mmWave line spacings of 123 GHz and 308 GHz.
Altogether, our work lays the theoretical and experimental foundation for designing wireless terahertz modulators
with smaller footprints, enabling scalability, broad bandwidths, and discusses strategies to achieve higher efficiencies
through resonant nonlinear effects.

2 Results

Antenna-transmission line co-design rules

Similar to conventional modulators, for our wired modulator we need to ensure that the mmWave signal on-chip is
maximized, while optimizing the interaction with the optical carrier. The antenna design dictates the amount of

4
power which is collected from free space and delivered to the transmission line, while the transmission line design
would impact the modulation efficiency between the optical probe and the mmWave signal. Our theoretical model
provides the following design guidelines:
i) The transmission line: Our wireless modulator performance metrics requires the transmission line to maintain
a coherence length (Lcoh ) larger than the device length LT L , minimize mmWave propagation loss αΩ , maximize spatial
overlap Γeo between the optical and mmWave modes, and achieve strong mmWave mode confinement by reducing
the mode cross-section SΩ . However, unlike conventional modulators, our wireless modulator does not require the
impedance matching between the characteristic impedance of the transmission line ZT L and 50 ohms cables and
delivery circuitry.
ii) The antenna: The antenna must be optimized to collect the maximum amount of the incident mmWave from
the free space Einc . Upon the incidence, the antenna generates a voltage wave Va across its gap. The goal is to
maximize the frequency-dependent inverse antenna factor (IAF = Einc ).
Va
Additionally, we need to ensure impedance
matching between the antenna and the transmission line (Za = ZT L ) to accomplish maximum delivery of the collected
wave to the transmission line.

Wideband mmWave wireless modulator

To meet the design rules, we optimize two critical transmission line parameters, namely, the electrode gap (w =
3.3 µm), and the optical waveguide width (wwg = 1.5 µm) to achieve impedance matching, minimize the metal-
related absorption, and maximize the coherence length (full design spec in the Supplementary Information section 1).
We achieve an optical loss of 1 dB/mm, mode overlap factor of Γeo = 0.54, and coherence length of Lcoh = 5 mm
with the aforementioned values for the waveguide width and electrode gap. We note that the measured optical loss is
higher than state-of-the-art values [43], however we chose this narrower gap to achieve a small mode area and hence
improve the modulation, highlighting the importance of balancing these parameters rather than minimizing the loss
alone.
On the antenna’s side, we opt for a dipolar design as opposed to bow-tie antennas since its resonant behavior
will boost the IAF . However, this comes at the cost of a frequency-dependent impedance Za which complicates the
impedance matching with the transmission line leading to reflection of the incident mmWave signal at the transmission
line interface (see details in the Supplementary Information section 5). Our simulations confirm that an antenna design
with a length of 400 µm gives high values of IAF at frequency range 50-500 GHz and a moderate reflection coefficient
ra = 0.6. Further details in Supplementary Information section 2.1 and 8.
The devices are fabricated in an x-cut thin film lithium niobate substrate similar to our previous work [36, 37].
The transmission lines are LT L = 2 mm long which is limited by mmWave loss. We illuminate our devices with
free-space mmWave radiation from a frequency extender system equipped with horn antennas emitting a Gaussian
mmWave beam (Fig. 2 a). We investigate two scenarios: configuration A - short range (< 10 cm) to simulate signal
delivery in a constrained space such as inside a cryostat for quantum experiments and configuration B - medium
range (10 − 100 cm) for applications in open spaces, such as 6G communications (shown in Fig. 2 b). Details on the
fabrication and the experimental setup in the methods and Supplementary Information section 1.
First, we butt-couple 1550 nm continuous wave light to the chip through its edge (9 dB/facet). The polarization
of the light is aligned with the z-axis of lithium niobate (TE). To ensure the maximum modulation efficiency, and
(2)
benefiting from the largest component of the nonlinear susceptibility tensor χ333 , the antenna collects the free-space
radiation into the plane of TFLN (z-axis), and couples it to the transmission line generating an electric field parallel

5
 a Experimental setup b Top view
Si lens
RF source (fRF)
mmWave
mmWave source source
(fmmW = 9×fRF or 27×fRF) ~5 cm
Configuration A
Horn antenna

Laser mmWave radiation OSA TPX lens Configuration B

Dipole antenna mmWave
Transmission line 0.4 mm source
FPC
TFLN
TFLN chip ~0.55 m chip
c Configuration A d Configuration B
PmmW = +20dBm PmmW = +5dBm Sideband
WR9.0 band ratio
WR2.8 band

e f

g h

Fig. 2: Experimental results demonstrating the performance of wireless TFLN electro-optic modula-
tors. (a) Schematic of the setup: an RF source generating 10 dBm of continuous-wave microwave signal at frequencies
between 9–14 GHz drives a frequency chain multiplier (mmWave source) with a multiplication factor of 9 (WR9.0
band) or 27 (WR2.8 band). The resulting mmWave signal is emitted into free space via a horn antenna and illuminates
the photonic chip, which is simultaneously pumped by 1550 nm light from a laser. The output spectra are recorded
using an optical spectrum analyzer (OSA). FPC – fiber polarization controller. The plot also includes a photograph
of the measured device, which shows a TFLN rib waveguide along which we pattern the antenna-coupled transmis-
sion line of length LT L . (b) Top-view sketch of the experimental setup: Configuration A (short range) investigates
our modulators upon short-distance illumination. The mmWave beam is focused onto the chip using a single TPX
(polymethylpentene) lens, with a total distance of 5 cm from the horn antenna to the chip. Configuration B (medium
range) corresponds to a setup with an increased distance of 0.55 m, where the beam is first collimated by a TPX
lens and then focused by a second TPX lens. In both cases we attach a hyper-hemispherical silicon (Si) lens to the
back-side of the TFLN chip. (c) Measured optical spectra for various mmWave frequencies in Configuration A, offset
from the optical carrier frequency. (d) Measured optical spectra for various mmWave frequencies in Configuration
B, offset from the optical carrier frequency. (e) Sideband ratio per mW of incident mmWave power as a function of
frequency in Configuration A. (f) Sideband ratio per mW of incident mmWave power as a function of frequency in
Configuration B. (g) Coupling efficiency (η) as a function of frequency for Configuration A, calculated from Eq.1,
with a linear fit shown as a dashed line. (h) Coupling efficiency (η) as a function of frequency for Configuration B,
with linear fit shown as a dashed line.

to the polarization optical field. In the frequency domain, the phase modulation (three-wave mixing) between the
optical carrier and the mm-Wave generates two optical sidebands. We quantify the modulator’s efficiency with the

6
side band ratio (SBR) at a given mmWave power incident onto the chip PmmW
f ree−space
, which is the ratio between the
optical power in each sideband and the power of the optical signal. SBR depends on the experimental parameters as
follows:
(χ(2) · ωo )2 · Γ2eo · L2T L · P M 2
SBR = f ree−space
· ηPmmW (1)
2 · n2o · nΩ · c30 · ε0 · SΩ
(2)
where χ(2) = χ333 , ωo is the optical angular frequency, no is the effective refractive index of the optical mode,
c0 is the speed of light in vacuum and ε0 is the vacuum permittivity. P M = ei∆k̃LT L −1
∆k̃LT L
is the phase-matching
function quantifying any propagation phase acquired between the interacting fields along the transmission line’s
2πfmmW
length ∆k̃ = c0 (nΩ − ng ) + i α2Ω , η is the coupling efficiency of mmWave power into the transmission line being
proportional to the square of IAF, and PmmW
f ree−space
is the power of the mmWave beam in free-space. Using these
definitions, the corresponding on-chip power is PmmW
on−chip
= η · PmmW
f ree−space
(complete derivation in Supplementary
Information section 2).
Our mmWave source provides 100 mW in the WR9.0 frequency range (82-125 GHz) and 5 mW in the WR2.8
frequency range (240-380 GHz) through a horn antenna which is coupled into the chip using a backside mounted
silicon lens. The modulated optical signal is collected using a lensed fiber, and then visualized on an optical spectrum
analyzer (Fig. 2 a-b, details in methods). Our wireless electro-optic modulators support efficient formation of side-
bands across the entire incident mmWave range (Fig. 2 c-d). To gain quantitative insight into the efficiency of
our broadband modulators and eliminate the effect of the variation in mmWave power from the source, we analyze
the frequency-dependent efficiency of the modulator by reporting the sideband ratio per unit of milliwatt incident
mmWave power (Fig. 2 e-f). In both configurations the antenna supports a spectrally flat sideband ratio across both
bands. We extract values of 10−4 /mW and 3 · 10−5 /mW for configuration A and configuration B, respectively. Using
Eq. 1 we can estimate the amount of mmWave power coupled from free-space to the chip to be 1-8% across the WR9.0
band and 2-6% across the WR2.8 band, respectively (Fig. 2 g-h). The linear frequency dependence of these values is
related to the frequency-dependent beam waist of the Gaussian
q mmWave beam generated by the horn antenna. We
on−chip
calculate a half-wave voltage-length product Vπ · LT L = 2ZT L SBR · LT L of 3.4 V·cm in WR9.0 and 3.65 V·cm
PmmW

at WR2.8 band, comparable to previous studies using wired mmWave electro-optic modulators [8].

Photonics-enabled characterization of mmWave transmission lines

The performance of our wireless modulators strongly depends on their mmWave loss, impedance and the transmission
line’s refractive index. At RF frequencies, these properties are routinely measured by performing a scattering matrix
analysis using a vector network analyzer. For optimization purposes, it’s essential to measure these values for our
devices, but such instruments are expensive in the mmWaves, and their calibration requires a wired signal delivery.
We propose a photonics-enabled technique to extract the transmission line’s parameters. Our method involves
performing two experiments, as shown in Fig. 3 a and b. The direction of propagation of the optical signal is reversed
between the two experiments, enabling interaction with the mmWave signal in either an optics-forward or an optics-
backward configuration. In both cases, the mmWave travels forward along the transmission line, gets reflected at
its end, and the returns. With phase-matching and sufficient single-pass mmWave losses (achieved by using long
transmission lines), the mmWave signal is attenuated during its forward propagation. This makes the modulation
more efficient in the optics-forward configuration compared to the optics-backward configuration where the added
attenuation of the mmWave signal during forward propagation leads to a smaller side-band ratio, see Fig. 3 c for
a transmission line length of LT L = 0.5 mm. In addition, for short transmission line lengths LT L , phase-matching

7
 a Optics-forward pumping b Optics-backward pumping
Forward propagating Forward propagating
AΩ AΩ

AΩe-ꭤ L
Ω TL
AΩe-ꭤ L
Ω TL

Backward propagating Backward propagating

c d

e f g h

Fig. 3: Photonics-enabled characterization of loss and refractive index of mmWave transmission lines.
(a) Sketch of the experimental setup for the optics-forward pumping scheme. The optical signal is coupled from the
left to the chip. The mmWave signal couples through the antenna to the transmission line, and has an amplitude
AΩ . It hence co-propagates with the optical signal in forward direction. As the mmWave signal propagates along the
transmission line, it gets attenuated due to losses αΩ . After reaching the end of the transmission line, the mmWave
signal is reflected with approximately unity efficiency. The amplitude of the back-reflected wave is then AΩ e−αΩ LT L .
This counter-propagating mmWave signal also interacts with the optical signal, however with reduced efficiency since
counter-propagating. Overall, the two processes (co- and counter-propagating optical and mmWave signals) lead to
the formation of sidebands in the optical domain (shown as dark green lines in the output spectra). (b) Sketch of the
experimental setup for the optics-backward pumping scheme. In this case the direction of propagation of the optical
signal is swapped compared to the optics-forward scheme. The optical signal interacts with the forward-propagating
mmWave signal in a counter-propagating configuration, whereas it interacts with the reflected mmWave signal in
co-propagating configuration. For long transmission lines, the co-propagating configuration is better phase-matched,
which results in weaker sidebands (shown as light green lines in the output spectra) since the back-reflected mmWave
signal is attenuated due to mmWave loss. (c) Normalized sideband ratio for the optics-forward (dark green) and optics-
backward (light green) configurations. We observe a lower sideband ratio for the optics-backward configuration, which
allows estimating the transmission line losses αΩ . There is also a clear frequency dependency, enabling the extraction
of the antenna’s reflection coefficient ra , see text. (d) Ratio of optics-forward to optics-backward sideband ratios,
shown in dB (gray points), along with the fitted curve (blue). The clear frequency dependency facilitates extracting
the refractive index of the mmWave mode propagating along the transmission line, see text. (e) Comparison between
simulated and fitted values of the refractive indices. (f) Comparison between simulated and fitted values of the
mmWave losses. (g) Comparison between simulated and fitted absolute values of the antenna-transmission line
reflection coefficient ra . (h) Comparison between the simulated and fitted phase of the reflection coefficient.

occurs for both co-propagating and counter-propagating mmWave and optical signals, and the total modulation is a
sum of these two contributions. This interference introduces the sinusoidal modulation in the ratio of the sideband-
ratio Fig. 3 c. Relating the sideband ratios of these two measurements (shown in Fig. 3 d) allows extracting the
transmission line’s loss αΩ , shown in Fig. 3 f and the transmission line’s refractive index, shown in Fig. 3 e. The
extracted values agree well with the simulated values.
With experimental values for loss and refractive index now determined, a remaining parameter is the reflection

8
coefficient at the antenna-transmission line interface. Given the impedance mismatch between the antenna and the
transmission line, a weak cavity is formed by the transmission line, with a reflectivity dictated by the antenna on
one end and the open circuit termination on the other end, leading to fringes in the side-band ratio in Fig. 3 b. By
comparing various LT L =0.25, 0.5 and 1 mm we extract complex reflection coefficient of the antenna ra , and by
exploiting the fact that reflection losses are unaffected by LT L , whereas propagation losses increase linearly with LT L .
We find a reflection coefficient ≈ 0.62 at 240-300 GHz and ≈ 0.73 around 100 GHz (see Fig. 3 g and h). Using these
deduced parameters, we can model the experimental data shown in Fig. 2 e-f by applying eq. 1, which shows good
agreement. Further details are provided in the methods and in the Supplementary Information section 5.
We note that the mmWave loss in our devices can be reduced by increasing the thickness of the transmission
line electrodes (see Supplementary Information section 4). Additionally, the overall efficiency of the system can be
further improved by enhancing the coupling between the THz wave and the transmission line through implementing
broadband antennas (see Supplementary Information section 8).

High-speed mmWave detection

In the context of a 6G communication scheme, a mmWave signal serves as the carrier, with data packets encoded
onto it via amplitude or phase modulation. This modulation process broadens the signal’s bandwidth. Specifically, it
extends the carrier’s frequency range by twice the modulation frequency. Consequently, detecting the full span of this
modulated mmWave signal requires a broadband device capable of covering the entire resulting frequency spectrum.
This bandwidth requirement also applies to other applications such as high-speed mmWave sensing.
To verify our modulator’s suitability for these applications, we now implement this device as a high-speed broad-
band mmWave detector. In this configuration, we amplitude modulate a 278.1 GHz carrier using a high-speed
electronic mixer. To investigate the electronic bandwidth of our detector, we apply frequencies spanning from 10
MHz to 6 GHz to the mixer by connecting a RF source to the intermediate frequency (IF) port of the mixer. A sketch
of the experimental setup is shown in Fig. 4 a.
In this configuration, the generated mmWave power is -21 dBm (vs +5 dBm in the previous experiments), limited
by the power handling of the electronic mixer (-11 dBm of IF power to maintain <1 dB compression level) and
its conversion loss (10 dB). Given the sideband ratio per mW of mmWave power (Fig. 2 e), we expect a sideband
ratio of approximately -67 dB (vs -38 dB in the experiments we discuss above). To overcome these challenging
power constraints, the optical signal is preamplified using an Erbium-Doped Fiber amplifier (EDFA), resulting in an
increased input optical power of +20 dBm before the chip, thereby increasing the sideband power. We note that this
frequency has been chosen arbitrarily and similar performance was observed for other carrier frequencies.
We investigate the bandwidth of our wireless phase modulators using configuration A (Fig. 2 b) while monitoring
the optical spectrum after the modulator on a high-resolution optical spectrum analyzer. For an amplitude-modulated
mmWave signal, we observe two peaks at frequencies fmmW − fmod and fmmW + fmod (Fig. 4 b). By sweeping the
modulation frequency from 10 MHz to 6 GHz (currently limited by the RF mixer), we observe that the separation
between the two peaks scales linearly with fmod . The large modulation bandwidth of our modulators enables the
amplitude of the sideband to remain relatively flat throughout the sweep making it suitable for applications such as
6G or high-speed mmWave/terahertz detection (Fig. 4 c).

9
 a Mixer LO input
Local oscillator (LO)
IF input f mmW=278.1 GHz
RF source (fmod)
Modulated output
Laser EDFA FPC High resolution OSA

TFLN chip

b c

Fig. 4: Experimental results on high-speed detection of wireless mmWave radiation. (a) A sketch of the
experimental setup used to characterize the response of our traveling wave electro-optic modulator under illumination
with an intensity-modulated mmWave signal oscillating at 278.1 GHz. The modulation signal is sinusoidal, oscillates
with a frequency fmod and is generated by an RF source. An electronic mixer imprints the modulation signal onto
the intensity of the mmWave signal. The mmWave is connected to the local oscillator (LO) port and the modulation
signal is connected to the intermediate frequency (IF) port. An optical signal is coupled through the edge of the chip
and the output spectrum is monitored on a high-resolution optical spectrum analyzer (OSA) after amplification with
an erbium doped fiber amplifier (EDFA) and controlling its polarization using fiber components (FPC). (b) Optical
spectrum at offset frequencies from the optical carrier, under ilumination with a mmWave signal modulated at 3 GHz.
Two optical tones are clearly visible, offset by fmmW from the optical carrier and separated by 2 · fmod , confirming
efficient intensity modulation of the mmWave. As expected, this is accompanied by the suppression of the fmmW
tone. The incident free-space mmWave power was 7.65 µW, the inset plot shows the zoomed region marked with a
black rectangle; (c) Normalized sideband ratio as the function of modulation frequency, showing a flat modulation
efficiency with frequency up to 6 GHz. Dashed lines represent ±3 dB bounds.

MmWave electro-optic frequency combs

Apart from traveling-wave electro-optic modulators, cavity-based modulators play a pivotal role in advancing mmWave
photonics. By enabling optical photons to make multiple round-trips within the cavity, these modulators significantly
enhance photon-photon interaction probability compared to single-pass configurations. In particular, cavity-based
modulation schemes enable applications such as electro-optic transduction [44] and ultra-sensitive field sensing [45]
even under weak mmWave illumination. Under strong mmWave illumination, the generated photons can interact
multiple times with the mmWave photons, leading to cascaded sideband generation and resulting in the formation of
an optical frequency comb.
Resonant electro-optic frequency comb sources can be driven by the multiples of the optical cavity’s FSR. In prac-
tice, this is not straightforward to do with conventional electronics due to technical challenges in realizing modulators
operating up to hundreds of gigahertz bandwidth. However, our wideband modulator provides access to a broad
range of frequencies. To achieve an electro-optic frequency comb we fabricate a racetrack resonator with a designed
free-spectral range of the cavity be 30.79 GHz. We place a 1.5 mm-long mmWave antenna-coupled transmission line in
one of the arms of the racetrack to phase modulate the light circulating in the cavity (Fig. 5 a). We use configuration
A to characterise our mmWave electro-optic comb source. Give that

• the frequency of the optical radiation matches the resonant frequency of the cavity, and

• the frequency of the mmWave radiation matches n × F SR , where n is an integer number and FSR is free

10
 a Free-space b f mmW= 4 • FSR
mmWave Optical resonant
radiation nonlinearity

χ(2)

Cavity modes
Input
f mmW= 10 • FSR
Racetrack resonator

0.4 mm
Output
TFLN chip Frequency

d e f

Fig. 5: Experimental results on wireless electro-optic comb generation. (a) A sketch of the setup. The
continuous-wave laser pumps the racetrack resonator, and the laser’s frequency matches to one of the racetrack’s
resonances. The chip is illuminated with the free-space mmWave radiation. The optical photon, passing multiple
times the transmission line, has a higher chance to generate an up- or down-converted sideband. In turn, if the
generated sideband becomes strong enough, it may interact with the mmWave photons again, leading to the cascading
effect and generation of the electro-optical comb. This system might be represented as a χ(2) in the optical resonator,
leading to an optically enhanced resonant nonlinear interaction. (b) A representation of the cascaded electro-optic
comb generation. The pumping laser (red line) pumps one of the optical resonances, and if the mmWave frequency
fmmW matches to an integer number times free-spectra range (FSR) of the racetrack, the generation process becomes
cascading leading to a larger electro-optic bandwidth. (c) the measured output spectra out of the chip by shining
mmWave radiation at frequencies of 123.2 GHz = 4 · FSR (upper plot, black line) and 307.9 GHz = 10 · FSR (bottom
plot, blue line) at maximal mmWave free-space powers; the comb slopes as the function of mmWave power at (d)
123.2 GHz and (e) 307.9 GHz; the upper x-axis indicating the calculated mmWave on-chip power; (f) calculated
single-photon coupling rate g0 for 307.9 GHz (blue curve) and 123.2 GHz (black curve); the dashed lines show the
mean value over the whole power range.

spectral range of the cavity,

the theoretical model shows that in the weak-pump regime, where electro-optic coupling strength (geo ) is smaller than
optical resonance linewidth (κ), the comb slope S [dB/GHz] is given by [46]:
 2 
10 4geo
S≈− log10 . (2)
fmmW κ2
√
Here, geo = g0 nmmW , with g0 being single-photon electro-optic coupling rate and nmmW is the number of mmWave
photons in the transmission line.
We design the device to be under-coupled and measure its intrinsic linewidth to be κ = 220 MHz (see Supplemen-
tary Information Section 6). We record the output spectra of the racetrack resonator under two distinct illumination
frequencies: 123.2 GHz, corresponding to 4×FSR, and 307.9 GHz, corresponding to 10×FSR of the ring (Fig. 5b).
The comb offers a characteristic slope of 0.05 dB/GHz (Fig. 5 c) and span a bandwidth of approximately 2 THz.

11
We further verify that the comb slope scales with the incident free-space mmWave power at both illumination fre-
quencies, showing a linear dependence consistent with Eq. 2 for both mmWave frequencies of 123.2 GHz as at 307.9
GHz (Fig. 5 d-e), see details in Supplementary Information section 7. However, we find that achieving the same slope
requires a lower power at 307.9 GHz than at 123.2 GHz (e.g. 2 dBm at 123.2 GHz versus -5.7 dBm at 307.9 GHz),
indicating that the efficiency of the comb generation per mW of mmWave power is larger at higher frequencies. We
attribute this higher efficiency to the increased single-photon electro-optic coupling rate at higher frequencies, which
√
scales as g0 ∝ fmmW [46]. The estimated single-photon electro-optic coupling rates g0 /2π from the experimental
data are 10.61 ± 1.11 kHz and 25.99 ± 3.18 kHz at 123.2 and 307.9 GHz, respectively (Fig. 5 f). These values are
approximately 10 times higher than the previous reports [15, 46] (see full analysis in Methods).

3 Conclusions and outlooks

In summary, we present a theoretical and experimental demonstration for a wireless electro-optic modulation scheme
on the thin-film lithium niobate platform. In this approach, the mmWave radiation is coupled directly to the chip via
an integrated antenna. This wireless coupling approach offers practical advantages eliminating the need for impedance-
matched transmission lines and avoiding the complexity and cost associated with high-frequency mmWave probes.
This wireless architecture offers a practical and scalable pathway for high-speed modulation in integrated photonic
systems. We developed experimental design guidelines for implementing wireless modulators across broad mmWave
frequency bands and introduced a photonics-based metrology technique to characterize key performance metrics,
including the refractive index, transmission line loss, and the complex reflection coefficient at the antenna–transmission
line interface. We demonstrated coupling efficiencies of up to 8% and showcased a path towards further improvement
through antenna engineering. In addition, we demonstrated a reflection coefficient of ≈ 0.62 in the 240-300 GHz
frequency range and ≈ 0.73 at frequencies around 100 GHz. This is particularly significant for the development of
on-chip mmWave/terahertz cavities for quantum optics and quantum computing applications.
Furthermore, combining lithium niobate properties and strong mmWave confinement enabled a nearly tenfold
enhancement in the single-photon electro-optic coupling rate g0 , compared to previously reported values. A key
challenge is reducing mmWave losses, for which we identify the use of thicker transmission lines as a promising
solution. Notably, this issue can be partially alleviated in superconducting circuits, where resistive losses are inherently
suppressed.
This novel architecture holds promise for broadening the scope and applicability of electro-optic modulators
across a wide range of disciplines. By eliminating the need for direct electrical contact and enabling efficient free-
space coupling, the wireless configuration not only simplifies system integration but also unlocks new opportunities
in environments where conventional wired approaches are impractical or lossy. In particular, this approach is ideally
suited for mmWave and terahertz sensing applications, where compact, high-bandwidth, and low-loss modulators are
essential for detecting weak electromagnetic fields. Perhaps most compellingly, this platform opens new frontiers in
quantum technologies. The ability to perform high-frequency electro-optic modulation without direct wiring could
become relevant for emerging qubit platforms [17]. In such systems, minimizing thermal load, wiring density, and
electromagnetic interference is critical, and the proposed wireless modulator offers a scalable path toward compact,
low-noise quantum interfaces. As the demand for integrated photonic systems continues to grow across both classical
and quantum domains, this wireless electro-optic architecture represents a foundational step toward enabling next-
generation high-frequency photonic technologies.

12
 Our results lay the foundation for compact, efficient, and scalable wireless electro-optic modulators, unlocking new
possibilities in high-frequency sensing, free-space communication, and quantum information science.

4 Methods

Fabrication

The chips are fabricated on 600 nm of X-cut lithium niobate, bonded to 4700 nm of thermally grown oxide on an
approximately 500 µm-thick double-side polished high-resistivity silicon substrate. The waveguides are patterned
using electron-beam lithography (Eliionix ELS-HS50) and Ma-N resist. These waveguides are then etched into the
LN layer using Ar+ ions, followed by annealing in an O2 environment to recover implantation damages and improve
the absorption loss of the platform [47]. Subsequently, we clad the devices with 800 nm of Inductively Coupled
Plasma - Chemical Vapor Deposition (Oxford Cobra), followed by another annealing. The electrodes are defined
using a self-aligned process, including patterning with PMMA resist, dry etching, and lift-off using the same resist.
The electrodes are deposited using electron-beam evaporation (Denton) with 15 nm of Ti and 300 nm of Au.

Experimental setup and measurements

The tunable Keysight N7776C laser generates a 1550 nm continuous wave 16 mW pump, which is then coupled
by lensed fiber (OZ Optics TSMJ-X-1550-9/125-0.25-7-2.5-14-2) to the lithium niobate waveguide on the chip from
the edge. The polarization is aligned to the TE mode of lithium niobate waveguide by a Thorlabs FPC561 Fiber
Polarization Controller. In the measurements (section 2) of wireless electro-optic modulator bandwidth of the main
text, there is also an EDFA (Erbium-Doped Fiber Amplifier) Nuphoton Technologies CW-C0-MR-30-20-FCA with a
gain in power of 10. For the outcoupling we use (OZ Optics TSMJ-X-1550-9/125-0.25-7-2.5-28-5).
The mmWave radiation is generated the following way: an Anritsu RF/Microwave Signal Generator MG362x1A
sends an RF signal (9-14 GHz) to the mmWave source (Virginia Diodes Signal Generator Extension module, WR9.0
+ WR2.8), which provides a 9x or 27x frequency up-conversion and emits up to 100 mW in WR9.0 range and up to
5 mW in WR2.8 range. The mmWave radiation is emitted into the free space via a horn antenna (Virginia Diodes
WR8.0CH antenna is used in the frequency range of 80-125 GHz with a beam waist at central frequency: 5̃.2 mm
and WR2.8DH antenna in the frequency range of 240-380 GHz, beam waist at central frequency: 1̃.9 mm). In the
configuration A of the text we use 1 inch aperture, 10 mm focal length TPX lens to tightly focus the mmWave beam
on the silicon lens, while in the configuration B we used a pair of TPX lenses: the first one, with an aperture of 1 inch
and focal length of 20mm, is used to form a collimated beam after the output of the horn antenna, and the second
one, with a large 2 inch aperture and 65mm focal length is used to focus the beam on the silicon lens. Then mmWave
radiation is collected by the on-chip antenna and coupled into the transmission line. Thus, the optical and THz modes
co-propagate and mix, creating sidebands detected by the OSA (Anritsu Optical Spectrum Analyzer MS9740B). For
the measurements in section 2 the high-resolution OSA (Apex AP2043B) was used. The Keysight MXG Vector Signal
Generator was used to generate the modulation microwave signal for the experiment in section 2. A balanced mixer
from Virginia Diodes (WR2.8BAMULP) was used to generate the modulated mmWave signal. The TPX lens from
Batop (LTA-D25.4-F10) was used to focus free-space mmWave radiation on the chip. However, in section 2, for the
measurements in configuration B, two lenses were used: first to collimate the mmWave beam (LTA-D25.4-F25), and
then the second to focus the radiation on the chip (LTA-D50-F65). The Hyperhemispheric Silicon Lens from Batop

13
(LSH-D12-T7.13) was fixed on the back of the chip using a self-designed chip holder.

Photonics-enabled characterisation of mmWave transmission lines

We implement a photonics-enabled technique to extract the transmission line’s parameters. Our method relies on
analyzing the difference in sideband ratio dSBR is given by the following formula:
 i∆k̃+ LT L 
  e −1
+ ei2kΩ LT L −αΩ LT L · ei∆k̃− LT L −1
SBRf orward ∆k̃ L ∆k̃− LT L
dSBR = 10 log10 = 20 · log10  −i∆k̃−+L T L  (3)
SBRbackward + ei2kΩ LT L −αΩ LT L · e−i∆k̃+ LT L −1
e T L −1
∆k̃− LT L ∆k̃+ LT L

which depends on the phase matching in co-propagating and counter propagating configuration, respectively, expressed
2πfmmW
through ∆k̃± = c0 (nΩ ∓ ng ) ± i α2Ω . These two terms depend, in turn, on the terahertz refractive index
nΩ = n0 + √ where n0 and n′ are the fitting parameters, and the mmWave loss αΩ . As we show in Supplementary
′
n
fmmW
Information 4, the main origin of the loss is the metallic loss caused by the finite surface impedance of the gold
electrodes. Therefore αΩ depends on fmmW and σAu and geometrical parameters of the transmission line according
to [48, 38, 39, 40]. We note that the second term of the expression for the mmWave refractive index originates from
the complex part of the surface impedance, as was explored in [39]. ng was kept to be 2.25 as measured in [36].
Fitting this formula to our experimental resuls helps retrieve the mmWave refractive index and propagation loss of
the transmission line. Our fitted value of gold conductance σAu
f it
= 1.91 · 107 S/m, notably lower than the bulk value
(4.1 · 107 S/m), consistent with prior observations of evaporated thin films [49].
We extract the reflection coefficient at the antenna-transmission line interface as follows. As derived in the
Supplementary Information 2, an impedance mismatch between the antenna and the transmission line will result
in the mmWave signal being reflected at the antenna-transmission line interface, resulting in a weak cavity effect.
The transmission line forms a resonator with a reflectivity dictated by the antenna on one end and the open circuit
termination on the other end. The resonator acts as a filter and only mmWave fields oscillating at frequencies
that match its cavity modes can efficiently couple to the resonator, leading in visible fringes in the side-band ratio.
The visibility of these fringes correlates with the reflection coefficient at the antenna-transmission line interface.
Consequently, comparing various LT L allows extracting the complex reflection coefficient of the antenna ra which
affects the fringe visibility, since reflection losses are unaffected by LT L , whereas propagation losses increase linearly
with LT L . According to our theoretical model, the reflection coefficient ra might be extracted from:

SBR P M 2 · L2T L · η
∝ 2 (4)
1 − ra e−αΩ LT L ei·4πfmmW nΩ LT L /c0
f ree−space
PmmW

We used three independent parameters of the fit: nΩ , ra , and a proportionality constant in the formula above. η was
taken as the linear dependence shown in Supplementary Information 5.

Calculation of the single-photon coupling rate g0 from the electro-optic comb slope

First, the electro-optic coupling rate geo was computed from the slope according to eq. 2. Then, the number of
mmWave photons nmmW in the traveling wave regime was estimated as follows [46]:
on−chip
PmmW
nmmW = (5)
h̄ · (2πfmmW )2
√
And then, g0 is estimated as geo / nmmW and plotted in Fig.5f. The errorbars for g0 are calculated from the
measurement error of the comb slope S.

14
 Data Availability The data generated in this study will be made available in the Zenodo database prior to
publication.

Code Availability The code used to plot the data within this paper will be made available in the Zenodo database
prior to publication.

References
[1] D. Tauber, B. Smith, D. Lewis, E. Muhigana, M. Nissov, D. Govan, J. Hu, Y. Zhou, J. Wang, W.-J. Jiang,
et al. “Role of coherent systems in the next DCI generation.” Journal of Lightwave Technology, 41(4):1139–1151
(2023).

[2] Y. Huang, Y. Shen, and J. Wang. “From Terahertz Imaging to Terahertz Wireless Communications.” Engineering,
22:106–124 (2023).

[3] M. Haziq, S. Falina, A. A. Manaf, H. Kawarada, and M. Syamsul. “Challenges and opportunities for high-power
and high-frequency AlGaN/GaN high-electron-mobility transistor (HEMT) applications: A review.” Microma-
chines, 13(12):2133 (2022).

[4] W. KOU, S. LIANG, H. ZHOU, Y. DONG, S. GONG, Z. YANG, and H. ZENG. “A review of terahertz sources
based on planar Schottky diodes.” Chinese Journal of Electronics, 31(3):467–487 (2022).

[5] T. Harter, C. Füllner, J. N. Kemal, S. Ummethala, J. L. Steinmann, M. Brosi, J. L. Hesler, E. Bründermann,

A.-S. Müller, W. Freude, et al. “Generalized Kramers–Kronig receiver for coherent terahertz communications.”
Nature Photonics, 14(10):601–606 (2020).

[6] S. Jia, L. Zhang, S. Wang, W. Li, M. Qiao, Z. Lu, N. M. Idrees, X. Pang, H. Hu, X. Zhang, et al. “2×
300 Gbit/s line rate PS-64QAM-OFDM THz photonic-wireless transmission.” Journal of Lightwave Technology,
38(17):4715–4721 (2020).

[7] F. Meng and N. Zhu. “An MSCL-based attenuator with ultralow insertion loss and intrinsic ESD-protection
for millimeter-wave and terahertz applications.” IEEE Transactions on Microwave Theory and Techniques,
71(1):240–249 (2022).

[8] C. Wang, D. Fang, J. Zhang, A. Kotz, G. Lihachev, M. Churaev, Z. Li, A. Schwarzenberger, X. Ou, C. Koos, et al.
“Ultrabroadband thin-film lithium tantalate modulator for high-speed communications.” Optica, 11(12):1614–
1620 (2024).

[9] Y. Horst, D. Moor, D. Chelladurai, T. Blatter, S. Fernandes, L. Kulmer, M. Baumann, H. Ibili, C. Funck,
K. Keller, et al. “Ultra-wideband MHz to THz plasmonic EO modulator.” Optica, 12(3):325–328 (2025).

[10] M. Zhang, C. Wang, P. Kharel, D. Zhu, and M. Lončar. “Integrated lithium niobate electro-optic modulators:
when performance meets scalability.” Optica, 8(5):652–667 (2021).

[11] S. Zhu, Y. Zhang, J. Feng, Y. Wang, K. Zhai, H. Feng, E. Y. B. Pun, N. H. Zhu, and C. Wang. “Integrated
lithium niobate photonic millimetre-wave radar.” Nature Photonics, pages 1–8 (2025).

15
[12] Y. Yamaguchi, P. T. Dat, S. Takano, M. Motoya, S. Hirata, Y. Kataoka, J. Ichikawa, R. Shimizu, N. Yamamoto,
K. Akahane, et al. “Advanced optical modulators for sub-THz-to-optical signal conversion.” IEEE Journal of
Selected Topics in Quantum Electronics, 29(5: Terahertz Photonics):1–8 (2023).

[13] N. Couture, F. Bouchard, A. Sit, G. Thekkadath, D. England, P. J. Bustard, and B. J. Sussman. “Terahertz
electro-optic modulation of single photons.” arXiv preprint arXiv:2503.19667 (2025).

[14] G. Santamaría Botello, F. Sedlmeir, A. Rueda, K. A. Abdalmalak, E. R. Brown, G. Leuchs, S. Preu, D. Segovia-
Vargas, D. V. Strekalov, L. E. García Muñoz, et al. “Sensitivity limits of millimeter-wave photonic radiometers
based on efficient electro-optic upconverters.” Optica, 5(10):1210–1219 (2018).

[15] K. K. Multani, J. F. Herrmann, E. A. Nanni, and A. H. Safavi-Naeini. “Integrated sub-terahertz cavity electro-
optic transduction.” arXiv preprint arXiv:2504.01920 (2025).

[16] L. Yi, Y. Li, and T. Nagatsuma. “Photonic radar for 3D imaging: From millimeter to terahertz waves.” IEEE
Journal of Selected Topics in Quantum Electronics, 29(5: Terahertz Photonics):1–14 (2023).

[17] A. Anferov, F. Wan, S. P. Harvey, J. Simon, and D. I. Schuster. “A Millimeter-Wave Superconducting Qubit.”
arXiv preprint arXiv:2411.11170 (2024).

[18] A. Kumar, A. Suleymanzade, M. Stone, L. Taneja, A. Anferov, D. I. Schuster, and J. Simon. “Quantum-enabled
millimetre wave to optical transduction using neutral atoms.” 615(7953):614–619.

[19] A. Suleymanzade, A. Anferov, M. Stone, R. K. Naik, A. Oriani, J. Simon, and D. Schuster. “A tunable high-Q
millimeter wave cavity for hybrid circuit and cavity QED experiments.” Applied Physics Letters, 116(10):104001
(2020).

[20] S. Rajabali and I.-C. Benea-Chelmus. “Present and future of terahertz integrated photonic devices.” APL
Photonics, 8(8):080901 (2023).

[21] C. Wang, M. Zhang, B. Stern, M. Lipson, and M. Lončar. “Nanophotonic lithium niobate electro-optic modula-
tors.” Optics express, 26(2):1547–1555 (2018).

[22] A. J. Mercante, S. Shi, P. Yao, L. Xie, R. M. Weikle, and D. W. Prather. “Thin film lithium niobate electro-optic
modulator with terahertz operating bandwidth.” Optics express, 26(11):14810–14816 (2018).

[23] M. Xu, M. He, H. Zhang, J. Jian, Y. Pan, X. Liu, L. Chen, X. Meng, H. Chen, Z. Li, et al. “High-performance
coherent optical modulators based on thin-film lithium niobate platform.” Nature communications, 11(1):3911
(2020).

[24] S. H. Badri, M. V. Kotlyar, R. Das, Y. Arafat, O. Moynihan, B. Corbett, L. O’Faolain, and S. Ghosh. “Com-
pact modulators on silicon nitride waveguide platform via micro-transfer printing of thin-film lithium niobate.”
15(1):11681.

[25] P. Zhang, H. Huang, Y. Jiang, X. Han, H. Xiao, A. Frigg, T. G. Nguyen, A. Boes, G. Ren, Y. Su, et al. “High-
speed electro-optic modulator based on silicon nitride loaded lithium niobate on an insulator platform.” Optics
letters, 46(23):5986–5989 (2021).

16
[26] Y. Salamin, W. Heni, C. Haffner, Y. Fedoryshyn, C. Hoessbacher, R. Bonjour, M. Zahner, D. Hillerkuss,
P. Leuchtmann, D. L. Elder, et al. “Direct conversion of free space millimeter waves to optical domain by
plasmonic modulator antenna.” Nano letters, 15(12):8342–8346 (2015).

[27] Y. Salamin, I.-C. Benea-Chelmus, Y. Fedoryshyn, W. Heni, D. L. Elder, L. R. Dalton, J. Faist, and J. Leuthold.
“Compact and ultra-efficient broadband plasmonic terahertz field detector.” Nature communications, 10(1):5550
(2019).

[28] J. D. Witmer, T. P. McKenna, P. Arrangoiz-Arriola, R. Van Laer, E. A. Wollack, F. Lin, A. K. Jen, J. Luo,
and A. H. Safavi-Naeini. “A silicon-organic hybrid platform for quantum microwave-to-optical transduction.”
Quantum Science and Technology, 5(3):034004 (2020).

[29] P. Gräupner, J. Pommier, A. Cachard, and J. Coutaz. “Electro-optical effect in aluminum nitride waveguides.”
Journal of applied physics, 71(9):4136–4139 (1992).

[30] D. Chelladurai, M. Kohli, J. Winiger, D. Moor, A. Messner, Y. Fedoryshyn, M. Eleraky, Y. Liu, H. Wang,
and J. Leuthold. “Barium titanate and lithium niobate permittivity and Pockels coefficients from megahertz to
sub-terahertz frequencies.” Nature Materials, pages 1–8 (2025).

[31] A. Majkić, U. Puc, A. Franke, R. Kirste, R. Collazo, Z. Sitar, and M. Zgonik. “Optical properties of aluminum
nitride single crystals in the THz region.” Optical Materials Express, 5(10):2106–2111 (2015).

[32] F. Wan, J. Han, and Z. Zhu. “Dielectric response in ferroelectric BaTiO3.” Physics Letters A, 372(12):2137–2140
(2008).

[33] M. Wang, Y. Chen, S. Zhang, L. Dong, H. Yao, H. Xu, K. Chen, and J. Wu. “Perspectives of thin-film
lithium niobate and electro-optic polymers for high-performance electro-optic modulation.” Journal of Materials
Chemistry C, 11(33):11107–11122 (2023).

[34] I.-C. Benea-Chelmus, M. L. Meretska, D. L. Elder, M. Tamagnone, L. R. Dalton, and F. Capasso. “Electro-optic
spatial light modulator from an engineered organic layer.” Nature communications, 12(1):5928 (2021).

[35] A. Herter, A. Shams-Ansari, F. F. Settembrini, H. K. Warner, J. Faist, M. Lončar, and I.-C. Benea-Chelmus. “Ter-
ahertz waveform synthesis in integrated thin-film lithium niobate platform.” Nature Communications, 14(1):11
(2023).

[36] Y. Lampert, A. Shams-Ansari, A. Gaier, A. Tomasino, S. Rajabali, L. Magalhaes, M. Lončar, and I.-C. Benea-
Chelmus. “Photonics-integrated terahertz transmission lines.” arXiv preprint arXiv:2406.15651 (2024).

[37] A. Tomasino, A. Shams-Ansari, M. Lončar, and I.-C. Benea-Chelmus. “Large-area photonic circuits for terahertz
detection and beam profiling.” arXiv preprint arXiv:2410.20407 (2024).

[38] W. Gallagher, C.-C. Chi, I. Duling III, D. Grischkowsky, N. Halas, M. Ketchen, and A. Kleinsasser. “Sub-
picosecond optoelectronic study of resistive and superconductive transmission lines.” Applied physics letters,
50(6):350–352 (1987).

[39] U. D. Keil, D. R. Dykaar, A. Levi, R. F. Kopf, L. Pfeiffer, S. B. Darack, and K. West. “High-speed coplanar
transmission lines.” IEEE journal of quantum electronics, 28(10):2333–2342 (1992).

17
[40] F. Martin. Artificial transmission lines for RF and microwave applications. John Wiley & Sons (2015).

[41] H. Murata, H. Yokohashi, S. Matsukawa, M. Sato, M. Onizawa, and S. Kurokawa. “Antenna-coupled electrode
electro-optic modulator for 5G mobile applications.” IEEE Journal of Microwaves, 1(4):902–907 (2021).

[42] H. Murata. “Millimeter-wave-band electro-optic modulators using antenna-coupled electrodes for microwave
photonic applications.” Journal of Lightwave Technology, 38(19):5485–5491 (2020).

[43] C. Wang, M. Zhang, X. Chen, M. Bertrand, A. Shams-Ansari, S. Chandrasekhar, P. Winzer, and M. Lončar.
“Integrated lithium niobate electro-optic modulators operating at CMOS-compatible voltages.” Nature,
562(7725):101–104 (2018).

[44] L. Fan, C.-L. Zou, R. Cheng, X. Guo, X. Han, Z. Gong, S. Wang, and H. X. Tang. “Superconducting cavity electro-
optics: A platform for coherent photon conversion between superconducting and photonic circuits.” Science
Advances, 4(8):eaar4994 (2018).

[45] X. Ma, Z. Cai, C. Zhuang, X. Liu, Z. Zhang, K. Liu, B. Cao, J. He, C. Yang, C. Bao, and R. Zeng. “Integrated mi-
crocavity electric field sensors using Pound-Drever-Hall detection.” Nature Communications, 15(1):1386 (2024).

[46] J. Zhang, C. Wang, C. Denney, J. Riemensberger, G. Lihachev, J. Hu, W. Kao, T. Blésin, N. Kuznetsov, Z. Li,
et al. “Ultrabroadband integrated electro-optic frequency comb in lithium tantalate.” Nature, pages 1–8 (2025).

[47] A. Shams-Ansari, G. Huang, L. He, Z. Li, J. Holzgrafe, M. Jankowski, M. Churaev, P. Kharel, R. Cheng, D. Zhu,
N. Sinclair, B. Desiatov, M. Zhang, T. J. Kippenberg, and M. Lončar. “Reduced material loss in thin-film lithium
niobate waveguides.” APL Photonics, 7(8):081301 (2022).

[48] G. Hasnain, A. Dienes, and J. Whinnery. “Dispersion of picosecond pulses in coplanar transmission lines.” IEEE
Transactions on Microwave Theory and Techniques, 34(6):738–741 (1986).

[49] Y. Lee, D. Kim, J. Jeong, J. Kim, V. Shmid, O. Korotchenkov, P. Vasa, Y.-M. Bahk, and D.-S. Kim. “Enhanced
terahertz conductivity in ultra-thin gold film deposited onto (3-mercaptopropyl) trimethoxysilane (MPTMS)-
coated Si substrates.” 9(1):15025.

Acknowledgments A.G., Y.L., and I.C.B.C. acknowledge funding from the European Union’s Horizon Europe
research and innovation programme under project MIRAQLS with grant agreement No. 101070700 and funding from
the Swiss National Science Foundation Grant No. №219406. S.R. acknowledges funding from the Hans Eggenberger
Foundation (independent research grant 2022, Switzerland) and the Swiss National Science Foundation (Postdoc
Mobility, grant number 214483). L.M. acknowledges funding from the Behring Foundation and CAPES-Fulbright.

Author contributions A.G. and I.C.B.C. conceptualized the project. A.G. and K.M. built the optical setup,
carried out the measurements, and performed the CST simulations. A.G. derived the theoretical description of the
modulation efficiency. Y.L. designed the layout of the photonic chip. J.L. helped with the measurements of the
electro-optic combs. A.S.-A., S.R., and L.M. fabricated the devices. A.G., A.S.-A. and I.C.B.C wrote the manuscript
with feedback from all authors. I.C.B.C.. and M.L. supervised this work.

Competing interests The authors declare no competing interests.

Disclaimer The views, opinions and/or findings expressed are those of the author and should not be interpreted
as representing the official views or policies of the Department of Defense or the U.S. Government.

18
 Corresponding authors Correspondence to Aleksei Gaier ([email protected]) or Ileana-Cristina Benea-
Chelmus ([email protected]).

19
 Supplementary information for
Large-area photonic circuits for terahertz detection and beam profiling
A. Gaier, K. Mamian, S. Rajabali, Y. Lampert, J. Liu, L. Magalhaes, A Shams-Ansari, M. Lončar, I.-C. Benea-Chelmus
Contents

1 Chip structure details 2

2 Theoretical model 3
2.1 Amplitude of the mmWave electric field coupled into transmission line from free-space . . . . . . . . . 5
2.2 Relation between voltage and electric field amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Loss model 8

4 Co- and counter-propagation mmWave radiation inside the transmission line 8

5 Measurements of the antenna-transmission line reflection coefficient 9

6 Linewidth of the optical microring resonance 10

7 Measurements of the coupling efficiency for the experiment with the ring resonators 11

8 Strategies on improving the coupling efficiency of the mmWave radiation to the chip 11

1 Chip structure details

Details on the geometry of the measured and simulated devices are given in SI Fig. 1, and all dimensions are provided
in SI Table 1. The refractive index of lithium niobate, silicon, and silicon oxide is calculated using the Lorentz model
from [refs].

thickness of the high resistivity silicon substrate hSi 500 µm

thickness of the silicon oxide layer hSiO2 4.7 µm
thickness of thin-film lithium niobate layer hT F 300 nm
thickness of the gold layer hAu 300 nm
height of the lithium niobate waveguide hwg 600 nm
thickness of the silicon oxide cladding hclad 1 µm
width of the lithium niobate waveguide wwg 1.5 µm
etching angle of the lithium niobate waveguide θwg 60◦
distance between the electrodes w 3.3 µm
width of the antenna wa 5 µm
width of the gold layer inside the transmission line wT L 3.5 µm
length of the transmission line LT L 0.25-2 mm
length of the dipole antenna La 400 µm

Tab. 1: Dimensions of the simulated and fabricated devices

2
 a b w c
wTL
wwg hAu hclad
La/2 Cross section θwg hwg
hTF

hSiO2
LTL

hSi
wa
LiNbO3 Si Au SiO2

Fig. 1: (a) Top view of the dipole antenna devices showing the transmission line with length LT L . (b) Cross-section
of the fabricated chips. (c) Simulated 2-D profiles of the optical and mmWave modes.

2 Theoretical model
In this chapter, we developed a theoretical description of the interaction of mmWave radiation coupled to the trans-
mission line. Namely, due to the nonlinear mixing of optical mode (at frequency ωp ) with mmWave field (at frequency
Ω), the sidebands at frequencies ωp ± Ω are generated.
We consider the linearly polarized optical fields along the z-axis, propagating along the y-axis. Then, the electric
field might be written as follows:
1
Eopt (x, y, z, t) = A(y)gopt (x, z)ei(ky−ωt) + c.c. (1)
2
where A(y) is the amplitude of the electric field, k = no cω0 is the propagation constant of the optical eigenmode with
the effective refractive index no and speed of light in vacuum c0 , gopt (x, z) is a 2D field profile of the eigenmode of the
RR
waveguide, normalized in the following way: |gopt |2 (x, z)dxdz = Sopt , where Sopt is the effective mode area. We
can assume that the nonlinear interaction is weak, which implies the non-depletion case of the optical probe, which
means that the probe field Ep and the sideband ESB could be described as follows:
1
Ep (x, y, z, t) = Ap gopt (x, z)ei(kp y−ωp t) + c.c. (2)
2
here we neglect the probe depletion due to the nonlinear interaction; note that Ap = const with respect to y, where
the subscript "p" stands for the probe mode.

(±) 1
ESB (x, y, z, t) = A (±) (y)gopt (x, z)ei(kSB(±) y−ωSB(±) t) + c.c. (3)
2 SB
where ωSB (±) = ωp ± Ω, subscript "SB" stands for sideband mode, supscript (±) stands for the upconverted (+) and
downconverted (-) sideband modes. The mmWave electric field (also linearly polarized along z-axis) can be described
as follows:
1 αΩ 1 αΩ
EΩ (x, y, z, t) = AΩ gΩ (x, z) ei(kΩ y−Ωt) e− 2 y + A− gΩ (x, z) e−i(kΩ y+Ωt) e 2 y + c.c. (4)
2 2 Ω
where kΩ = nΩ cΩ0 is the wave vector of the RF wave with an effective refractive index nΩ , αΩ is the propagation
loss of the mmWave radiation, and AΩ and A−
ω are the amplitudes of the mmWave waves that co- and counter-

propagate with an optical waves. Here we neglect the depletion of the mmWave amplitude due to the nonlinear
interaction. The 2D profile of the mmWave mode is given by gΩ function, which is normalized in the same way as
RR
gopt : |gΩ |2 (x, z)dxdz = SΩ .
The evolution of the optical electric field E = Ep + ESB + + ESB − is given by the nonlinear wave equation:
 
1 ∂2 1 ∂2

∇2 − 2 2 E(x, y, z, t) = 2 2
P L (x, y, z, t) + P N L (x, y, z, t) (5)
c0 ∂t ε0 c0 ∂t

3
 Rt
where P L (x, y, z, t) = ε0 −∞
χ(1) (x, y, z, t − t′ )E(x, y, z, t′ )dt′ is the linear part of the polarization with χ(1) being
2
linear susceptibility, and P N L (x, y, z, t) = ε0 χ(2) (x, y, z, t) · (E(x, y, z, t) + EΩ (x, y, z, t)) is the nonlinear part of the
polarization with χ(2) being second-order nonlinear susceptibility. To simplify the expression for linear polarization,
assuming the homogeneity of the material properties along the y-axis and using the properties of the Fourier transform,
one may obtain:
Z t
P L (x, y, z, t) = ε0 χ(1) (x, z, t − t′ )E(x, y, z, t′ )dt′ =
−∞
1
ε0 χ(1) (x, z, ωp )Ap gopt (x, z) · ei(kp y−ωp t) +
2 (6)
1
ε0 χ(1) (x, z, ωSB + )ASB + (y)gopt (x, z) · ei(kSB+ y−ωSB+ t) +
2
1
ε0 χ(1) (x, z, ωSB − )ASB − (y)gopt (x, z) · ei(kSB− y−ωSB− t) + c.c.
2
R
Computing the second derivatives and using orthogonality of the complex exponents ei(ω−ω )t dt = 2πδ(ω − ω ′ ), we
′

can rewrite the equation 5 to describe the evolution of the (+) sideband:
 2 
2 ωSB + 2
∇ + n (x, z, ωSB ) ASB + (y) · gopt (x, z) · eikSB+ y =
+
c20
  (7)
2χ(2) (x, z)ωSB
2
+ − 2Ω y i(kp +kΩ )y
α
− 2Ω i(kp −kΩ )y
α
− A g (x, z)g Ω (x, z) A Ω e e + A Ω e e
c20
p opt

∂2
Now, we can split the nabla operator into longitudinal and transversal components: ∇2 = ∆2⊥ + ∂y 2 , so that:
 2   2 !
2 ωSB + 2 ∂2 ωSB +
∇ + n (x, z, ωSB + ) gopt (x, z) = + no gopt (x, z) (8)
c20 ∂y 2 c0

Now, by multiplying both sides of the equation by gopt (x, z) and integrating over the (x, z) coordinates, one may
obtain:
 2   
∂2 ωSB + 2 2χ(2) ωSB
2
+ − 2Ω y i(kp +kΩ )y
α
− 2Ω i(kp −kΩ )y
α
+ n A + (y) · e SB
ik + y
= − A Γ A Ω e e + A Ω e e (9)
∂y 2 c20 c20
o SB p eo

where we introduced the spatial overlap factor as follows:

RR 2
gopt (x, z) · gΩ (x, z)dxdz
Γeo = LNRR (10)
g 2 (x, z)dxdz
x,z opt

where the integral in the nominator is taken over the nonlinear medium (in our case, lithium niobate waveguide), and
the area of the integral in the denominator is the whole XZ plane, and we introduced effective nonlinear susceptibility
(2)
χ(2) = χ333 . Calculating the second derivatives in the left part of the equation gives:
 2 2   2 
∂ ωSB + 2 ∂ ASB + ∂ASB +
+ n A + (y) · e ikSB + y
=e ikSB + y
· + 2ik + (11)
∂y 2 c20 ∂y 2
o SB SB
∂y

Substituting this result into the nonlinear wave equation gives:

∂ 2 ASB + ∂ASB + 2χ(2) ωSB

2  αΩ αΩ

+
2
+ 2ikSB + =− 2 Ap Γeo AΩ e− 2 y ei∆k̃+ y + A−
Ωe
2 ei∆k̃− y (12)
∂y ∂y c0
2πfmmW
where we introduced ∆k̃± = c0 (nΩ ∓ ng ) ± i α2Ω , with the group index of the optical mode ng . If now we apply
∂ 2 ASB +
the slowly varying envelope approximation assuming | | for eq. 12:
∂ASB +
∂y 2 | ≪ |2kSB + ∂y

∂ASB + χ(2) ωSB +  αΩ αΩ


=i Ap Γeo · AΩ e− 2 y ei∆k̃+ y + A−
Ω e 2 ei∆k̃− y (13)
∂y c0 n o

4
 IAFP99P  D E




           
fmmW*+] fmmW*+]

Fig. 2: Simulated inverse antenna factors for the antenna design used in this work in two configurations: with (a)
and without (b) Si lens.

The solution of this equation at coordinate y=L, assuming that at y=0 the sideband mode is not populated, or in
other words that ASB + (y = 0) = 0, is given by:
!
χ(2) ωSB + ei∆k̃+ L − 1 ei∆k̃− L − 1
ASB + = Ap Γeo L · AΩ + A−
Ω (14)
no c0 ∆k̃+ L ∆k̃− L

This solution is valid only for small modulation, i.e., |ASB + | ≪ |Ap |. However, in practice, the coherence length

coh =
L− ≪ Lm at frequencies above 100 GHz, and therefore the contribution of the second term is negligible:
π
∆k̃−

χ(2) ωSB + ei∆k̃L − 1

ASB + = Ap Γeo L · AΩ (15)
no c0 ∆k̃L

where we set ∆k̃+ = ∆k̃. The next step is to find the relation between the incident free-space mmWave electric field
Einc and the amplitude AΩ of mmWave electric field mode coupled to the transmission line. In the next section, we
will evaluate the amplitude of the mmWave electric field in the transmission line upon illumination with free-space
mmWave radiation.

2.1 Amplitude of the mmWave electric field coupled into transmission line from free-
space

The incident electric field Einc induces a voltage between antenna pads, which depends on the inverse antenna factor
(IAF):
Va = IAF · Einc (16)

The values of the inverse antenna factor for the antenna used in this work is shown in Fig.2. It has a clear
resonance at center frequency ≈300GHz, and reaching the value of 8 mV/(V/m). In this plot we compared two ways
of the antenna illumination: with and without silicon lens. Adding the silicon lens improved the antenna factor almost
ten folds. However, in practice, only if the input beam waist is much larger than the antenna aperture this formula
apply. In the case, for example, of antenna illumination with a gaussian beam with a waist smaller than the silicon
lens aperture, this value may differ from the simulated values and should be studied for the parameters of the input
beam.
The voltage generated by the antenna launches a voltage wave into the transmission line, and the relation between
antenna voltage and the transmission line is studied below.

5
2.2 Relation between voltage and electric field amplitudes

Consider the equivalent circuit for the devices with the antenna placed at the beginning of the transmission line shown
in Fig. 3, where the line is terminated with a reflection coefficient r at y = L. Antenna serves as a voltage generator
with Va and Za . The present circuit could be studied using an infinite series to represent the multiple bounces
from both ends of the transmission line, however for the sake of simplicity we will use the method of impedance
transformation described in [1]. The voltage and current in the line are given by the superposition of the waves
flowing in the positive and negative directions along the transmission line:

V (y) = V0+ e−γy + V0− eγy , (17)

1
I(y) = (V + e−γy − V0− eγy ), (18)
ZT L 0

The reflection coefficient r at the load (y = L) is the ratio of amplitudes of the left-traveling wave and the right-

Va γ, Z0
+
V0
Zin
Za V0
-

y
y=0 y=L

Fig. 3: Equivalent circuit scheme for the devices under study. The antenna serves as a voltage generator connected
in series to the transmission line. The reflection coefficient r from the end of the transmission line is close to 1, thus
an open circuit termination is schematically shown.

traveling one:

V0− eγL V0− 2γL

r= + −γL = e (19)
V0 e V0+

Putting this expression into the ones above gives:

V (y) = V0+ (e−γy + re−2γL eγy ), (20)

V0+ −γy
I(y) = (e − re−2γL eγy ), (21)
ZT L
V (0) 1 + re−2γL
Zin (y = 0) = = ZT L , (22)
I(0) 1 − re−2γL

where γ = αΩ
2 − ikΩ (αΩ > 0) is the complex propagation constant of the mmWave mode, ZT L is the characteristic
impedance of the line.
Since the phase-matching condition is satisfied primarily for the right-traveling wave, further we are interested
in V0+ . To find V0+ we note that on the one hand, V (0) = V0+ (1 + re−2γL ); on the other hand, from the antenna
side we can write based on the expression for impedances connected in series: V (0) = Va ZinZ+Z
in
a
. Equating these two
expressions for V (0) we get:
Zin 1
V0+ = Va (23)
Zin + Za 1 + re−2γL

6
It’s useful to rewrite eq. 23 in terms of reflection coefficients from both ends:
1+re
ZT L 1−re1−2γL
−2γL
Zin 1 ZT L 1−re 1
V0+ = Va = = = (24)
−2γL
V V
Zin + Za 1 + re−2γL 1 + re−2γL
a 1+re a 1+re−2γL
−2γL + Za −2γL + Za
−2γL
ZT L 1−re ZT L 1−re
ZT L ZT L
= Va = Va =
ZT L (1 + re−2γL ) + Za (1 − re−2γL ) ZT L + Za + re−2γL (ZT L − Za )
ZT L 1 Va 1 − ra
= Va =
ZT L + Za 1 − r · ra e −2γL 2 1 − ra · a · ei2kΩ L
where ra is the reflection coefficient from the antenna end of the transmission line: ra = Za +ZT L .
Za −ZT L
Thus,

Va 1 − ra Va
V0+ = · = · ta (25)
2 1 − ra · a · e i2kΩ L 2
where we introduced the transmission term (from the antenna to the transmission line) ta . One can see that this
formula contains the sum of an infinite geometric series corresponding to the multiple reflections from both ends of
the transmission line. The result represents a cavity where one mirror is the circuit termination, and the other is the
antenna-transmission line interface. Note that the Va , Za , ZT L , γ, r parameters in the results presented further were
simulated using the commercial software package CST Studio Suite. The simulated values of r are close to unity,
meaning that the end of the transmission line can be approximately considered an open circuit termination. To better
understand the physical meaning of the formula, let us consider two extreme cases for the transmission term.
1. Let ra = 1. This would happen if Za ≫ ZT L . Then V0+ = 0, meaning that all the mmWave field was reflected
at the "antenna - transmission line" interface, and no voltage entered the transmission line.
2. Let Za = ZT L , thus ra = 0 and the transmission line is matched to the antenna. Then V0+ = 2 .
Va

Fig. 8 demonstrates ra as a function of frequency for the dipole antenna and the transmission line used in this
work.
The right-traveling mmWave electric field amplitude AΩ might be found from the two forms of the expression for
the right-traveling power of in the cavity PmmW
on−chip
:
 
|V + |2 1 1
on−chip
PmmW = 0 Re = ε0 nΩ c0 |AΩ |2 SΩ (26)
2 ZT L 2
Our simulations, together with the analytical model from [2] have shown the imaginary part of the characteristic
 
impedance is negligible, so then Re ZT1 L ≈ ZT1 L . This gives the relation between the amplitude of the mmWave
field in the transmission line AΩ and the incident mmWave electric field Einc :
1
|AΩ | = IAF · |ta | · √ Einc (27)
2 ZT L ε0 nΩ c0 SΩ
Finally, with this expression, one may introduce the side-band ratio (SBR) defined for a given input mmWave
intensity
PSB + = SBR · Pp (28)

where PSB + , Pp are powers of the generated sideband and optical probe. Substituting eq. 27 into eq. 15 gives the
following expression for the side-band ratio:
(χ(2) · ωSB + )2
SBR = · IAF 2 · |ta |2 · P M 2 · Einc
2
· Γ2eo · L2 (29)
4· n2o · nΩ · c30 · ε0 · ZT L · SΩ

where we introduced the phase-matching function P M = ei∆k̃L −1

∆k̃L
describing the phase-matching of optical and
IAF 2 ·|ta |2 Einc
2
mmWave modes, as well as the loss of the mmWave mode. By denoting PmmW
on−chip
= 2ZT L = ηPmmW
f ree−space
we
get:
(χ(2) · ωo )2 · Γ2eo · L2T L · P M 2
SBR = f ree−space
· ηPmmW (30)
2 · n2o · nΩ · c30 · ε0 · SΩ

7
where we introduced the coupling efficiency coefficient η of mmWave radiation from free-space to the transmission
line and took into account that the probe frequency ωo = ωp ≈ ωSB + , and L = LT L .

3 Loss model
The analytical formulas for the various loss terms can be found in [2]. The main contributions to the loss are
the conductor and radiative losses αc and αrad respectively. By introducing the frequency-dependent skin depth
1


δ=√ and surface impedance ZS = 1+i
σδ coth (1 + i) δ
hAu
, the conductive losses are given by:
σ·fmmW ·µ·π

 
Zs
αc = g · Re dB/m (31)
ZT L

where the expression for the proportionality coefficient g is given by:

  4πwT L
P′ w
1.25
π ln( hAu ) + 1 + 1.25hAu

g = 17.34 · · 1+ · 
πwT L
2 (32)
πw wT L 2wT L
1+ w + 1.25πh
πw
Au
· 1 + ln( 4πw
hAu )
TL

where w, wT L , hAu are the parameters as presented in section 1, and P ′ is:

  2

 √ 2k K(k)
K (k) , for 0 ≤ k ≤ 0.707
P ′ = (1− 1−k )(1−k )
2 3/4 ′
(33)
 1 √
 ,
(1−k) k
for 0.707 ≤ k ≤ 1
√
with k = w+2wT L ,
w
K(k) being a complete elliptic integral of the first kind, and K ′ (k) = K( 1 − k 2 ). As shown in
the same work, the radiative losses are proportional to the third power of frequency:
√  2
53 − 8 nΩ n2Ω (w + 2wT L )2 3
αrad = π · 1− 2 · 3 ′ f (34)
2 nsub nsub c0 K (k)K(k) mmW

where nsub ≈ 3.41 is the substrate’s refractive index (silicon). Another contribution to the loss, a tangent loss αt , can
be simulated using, for example, CST Studio Software by simulating the 2D mode profile and relating the imaginary
part of the effective refractive index κΩ as:
4πκΩ fmmW
αt = (35)
c0
Combining all the above mentioned contributions of the loss we define the total loss αΩ = αc + αrad + αt and plot
these various contributions in Fig.4. As it is clearly seen, the most dominant contribution comes from the conductor
losses. One of the strategies on how to minimize this loss is to increase the metal thickness hAu . We plot the losses for
the various metal thicknesses hAu in Fig.4. One may notice that the losses can be reduced significantly by increasing
the metal thickness. The fabrication of the thicker electrodes might be challenging, however, it has been already
demonstrated in [3].

4 Co- and counter-propagation mmWave radiation inside the transmis-

sion line
As it was described above, there are two waves propagating in the transmission line with the voltage amplitudes of
V0+ and V0− , according to eq. 17. These voltage amplitudes correspond to the forward propagating electric field
wave with an amplitude AΩ and backward propagating wave A−
Ω . In the case of the short transmission lines, whose

length LT L are comparable to the coherence lengths of the backward propagating wave π
, the contribution of
|∆k̃− |

8
    
D E F G

radG%PP

G%PP
cG%PP
hAuQP
tG%PP
 
  

   
 
 
   
H I J K

radG%PP

G%PP
cG%PP
tG%PP

 
 
 
 
 
                   
fmmW*+] fmmW*+] fmmW*+] fmmW*+]

Fig. 4: Simulated loss for various electrode thicknesses and metal properties. (a, e) Tangent, (b, f)
radiative, (c, g) conductor and (d, h) total loss versus mmWave frequency. Upper row: σ = 1.91 · 107 S/m (as
retrieved from the experimental data). Lower row: σ = 4.1 · 107 S/m (literature value of gold conductance).

the second term in eq.14 is not negligible. Therefore, the electric field amplitude of the generated sideband Aco
SB + ,

when the optical probe co-propagates with the forward propagating mmWave mode, in accordance with eq.14 and 19
is proportional to: !
ei∆k̃+ LT L − 1 ei∆k̃− LT L − 1
Aco
SB + ∝ AΩ + ei2kΩ LT L −αΩ LT L (36)
∆k̃+ LT L ∆k̃− LT L
If the optical probe is sent the other way around, so that it counter-propagates the forward propagating mmWave
mode, then its amplitude will be given by
!
e−i∆k̃− LT L − 1 e−i∆k̃+ LT L − 1
Acounter
SB + ∝ AΩ + ei2kΩ LT L −αΩ LT L (37)
∆k̃− LT L ∆k̃+ LT L

as now the phase-matching conditions will be the opposite for the forward-
 and backward-propagating
 mmWave waves.
2
|Aco |
Now, by introducing the difference sideband ratio dSBR = 10 · log10 SB +
counter
2 , we get:
A
SB +

 
ei∆k̃+ LT L −1
∆k̃+ LT L
+ ei2kΩ LT L −αΩ LT L · ei∆k̃− LT L −1
∆k̃− LT L
dSBR = 20 · log10   (38)
e−i∆k̃− LT L −1
∆k̃− LT L
+ ei2kΩ LT L −αΩ LT L · e−i∆k̃+ LT L −1
∆k̃+ LT L

5 Measurements of the antenna-transmission line reflection coefficient

We performed the measurements of the sideband ratio, shown in Fig.6 of four devices with various lengths: 0.25, 0.5,
1 and 2 mm. Since each time we place the silicon lens on the back of the chip, there might be a slight displacement
of the antenna with respect to the center of the silicon lens, for each set of measurements, we calculated the coupling
efficiency. As a reference, we used the device with a 2 mm transmission line length. The results for this experiment
are shown in Fig. 5, together with the linear fit of the coupling efficiency, which will be used later below.
Then, we performed the measurements of the three left devices with the transmission line lengths 0.25, 0.5 and 1
mm and the results are shown in the Fig.6a-c.
The fitted value of the raf it = 0.653e−i·0.085π is very close to the simulated one plotted in Fig.3 of the main text
both in amplitude and in phase. Another interesting finding is the dependence of on- and off-resonant SBR per
f ree−space
PmmW as a function of the transmission line’s length. We noted that it scales as LT L for the on-resonance case,

9
 





       
fmmW*+]

Fig. 5: Coupling efficiency of the mmWave free-space radiation to the chip performed right before the reflection
coefficient measurements at the antenna-transmission line interface.

a LTL=0.25mm b LTL=0.5mm c LTL=1mm

d e

~L2TL ~L TL

Fig. 6: The sideband ratio per PmmW

f ree−space
as a function of mmWave frequency and the device photograph with the
length of transmission line (a) 0.25 mm; (b) 0.5 mm; (c) 1 mm. The dependence of (d) off-resonance (min value
across the whole frequency range) and (e) on-resonance (the mean value in the range 270-280 GHz) sideband ratio
per PmmW
f ree−space
as a function of the transmission line length. Dashed lines indicate the fits.

while as L2T L for the off-resonance case in good agreement with eq.29. This may indicate that in on-resonance case,
1
the field enhancement coming from the term 2 compensates for the shorter interaction lengths.
|1−ra ·a·ei2kΩ L |
Then, we performed the same measurements across WR9.0 band, and plotted the results in Fig.3 of the main text.
The fitted value of the raf it = 0.729ei·0.464π is again very close to the simulated one plotted Fig.3 of the main text.
We used the CST Studio software to simulate the impedances of the antenna and the transmission line. These
values are shown in Fig.8

6 Linewidth of the optical microring resonance

We performed the transmission measurement of the ring resonator in order to get the linewidth of the pumped
resonance. We plot the transmission spectrum of the resonance we pumped for the experiments in the electro-optic
comb generation in Fig.9 as well with the statistics over multiple resonances. We pumped the resonance with the
linewidth of ≈220 MHz and this value was further used in the calculations.

10
 a LTL=0.5mm b LTL=1mm

Fig. 7: The sideband ratio per PmmW

f ree−space
as a function of mmWave frequency and the device photograph with the
length of transmission line (a) 0.5 mm; (b) 1 mm.

 
 D $QWHQQD
7UDQVPLVVLRQ
 E
Re(Z)2KP

Im(Z)2KP
OLQH 




 
 
                   
fmmW*+] fmmW*+]

Fig. 8: The simulated real (a) and imaginary parts (b) of the antenna and transmission line impedance.

7 Measurements of the coupling efficiency for the experiment with the

ring resonators
As described above, we measured the coupling efficiency each day prior to conducting the main experiments. For
these measurements, we used a test device consisting of a 2mm-long transmission line with a dipole antenna. The
measured coupling efficiencies are presented below in Fig.10.
We note the coupling efficiency at frequencies below 125 GHz because the TPX lens could not be used due to the
setup’s constraints and the limited space behind the chip.

8 Strategies on improving the coupling efficiency of the mmWave radi-

ation to the chip
As discussed above, the coupling efficiency depends mainly on two parameters: IAF and ra . Then, IAF might be
optimized in two different ways: by optimizing the parameters of the free-space gaussian mmWave beam (such as the
beam waist) illuminating the chip by developing the lens system, or by engineering the design of the antenna and using
other types of antennas, such as bow-tie [4], bull eye [5], LC [6] and others. In addition, the reflection coefficient at
the antenna-transmission line ra significantly affects the coupling efficiency. Broad- or narrowband coupling efficiency
optimization is considered depending on the application. In the first case, the ra must be minimized across the
1−ra
entire frequency range of interest, bringing the values of ta = 1−ra ·a·ei2kΩ L
to 1. However, in the case of narrowband
optimization, one may consider exploiting the resonant behavior of the term ta as ra → −1. By minimizing the
round-trip propagation loss a, one may benefit from the enhancement of the intra-cavity field at frequency where
arg(ra · ei2kΩ L ) = 2πm with m being an integer number. We demonstrate the possibility of engineering the ra
coefficient to be significantly low across a large frequency range by simulating the reflection coefficient of the bow-tie

11
 
D E


1XPEHURIRFFXUUHQFH
 

W


3RZHU


 ([SGDWD 
)LW

         
f fres*+] 0+]

Fig. 9: (a) The transmission spectrum of the pumped resonance and (b) linewidth statistics over 256 resonances







       
fmmW*+]

Fig. 10: Coupling efficiency of the mmWave free-space radiation to the chip performed right before electro-optic comb
measurements.

antenna with a width of 300 µm and a height of 400 µm, as presented in the Fig. 11.



D $QWHQQD



E 

F
7UDQVPLVVLRQ
Re(Z)2KP

Im(Z)2KP

 OLQH 


|ra|

 

 
  
              
fmmW*+] fmmW*+] fmmW*+]

Fig. 11: Simulation results for the bow-tie antenna. (a) Real part of the impedance; (b) Imaginary part of the
impedance; (c) reflection coefficient between the bow-tie antenna and the transmission line demonstrating flat and
low reflection coefficient ra across frequency range of 200-1000 GHz.

References
[1] D. M. Pozar. Microwave engineering; 3rd ed. Wiley, Hoboken, NJ (2005).

[2] U. D. Keil, D. R. Dykaar, A. Levi, R. F. Kopf, L. Pfeiffer, S. B. Darack, and K. West. “High-speed coplanar
transmission lines.” IEEE journal of quantum electronics, 28(10):2333–2342 (1992).

[3] C. Wang, D. Fang, J. Zhang, A. Kotz, G. Lihachev, M. Churaev, Z. Li, A. Schwarzenberger, X. Ou, C. Koos, et al.
“Ultrabroadband thin-film lithium tantalate modulator for high-speed communications.” Optica, 11(12):1614–
1620 (2024).

12
[4] I.-C. Benea-Chelmus, T. Zhu, F. F. Settembrini, C. Bonzon, E. Mavrona, D. L. Elder, W. Heni, J. Leuthold, L. R.
Dalton, and J. Faist. “Three-dimensional phase modulator at telecom wavelength acting as a terahertz detector
with an electro-optic bandwidth of 1.25 terahertz.” Acs Photonics, 5(4):1398–1403 (2018).

[5] M. Beruete, U. Beaskoetxea, M. Zehar, A. Agrawal, S. Liu, K. Blary, A. Chahadih, X.-L. Han, M. Navarro-Cía,
D. Etayo Salinas, A. Nahata, T. Akalin, and M. Sorolla Ayza. “Terahertz Corrugated and Bull’s-Eye Antennas.”
IEEE Transactions on Terahertz Science and Technology, 3(6):740–747 (2013).

[6] Y. Salamin, I.-C. Benea-Chelmus, Y. Fedoryshyn, W. Heni, D. L. Elder, L. R. Dalton, J. Faist, and J. Leuthold.
“Compact and ultra-efficient broadband plasmonic terahertz field detector.” Nature communications, 10(1):5550
(2019).

13