Traveling-Wave Optical Isolators

• Traveling-wave optical isolators are nonreciprocal photonic devices that use traveling perturbations (acoustic, RF, or magneto-optic) to break symmetry in light propagation.
• They deploy diverse mechanisms, such as Brillouin scattering, electro‐optic modulation, and nonlinear adiabatic conversion, to enable unidirectional transmission while suppressing backward signals.
• Design trade-offs focus on optimizing phase matching, drive power, bandwidth, and insertion loss to meet application-specific requirements in integrated photonic systems.

Traveling-wave optical isolators are nonreciprocal photonic components in which forward and backward light encounter different phase matching, mode conversion, interference, or resonant conditions because the device embeds a traveling perturbation, a traveling acoustic phonon, a traveling microwave-driven phase modulation, or a unilateral chirped quasi-phase-matching profile. Across recent literature, the class includes linear Brillouin-scattering–induced transparency in dielectric waveguide–resonator systems, electro-optic traveling-wave phase modulators on thin-film lithium niobate, electrically driven acoustic-wave devices in suspended silicon, gyrotropic coupled-waveguide systems, resonant magneto-optic tunneling structures for circular polarization, and nonlinear adiabatic frequency-conversion schemes with strong absorption at the generated wave (Kim et al., 2016, Yu et al., 2022, Dostart et al., 2018, Chao et al., 14 Feb 2025, Moccia et al., 2013, Rangelov et al., 2016).

1. Physical basis of traveling-wave isolation

A central feature of the traveling-wave optical isolator is that the interaction is direction-selective in momentum space, in spatiotemporal phase, or in both. In the Brillouin-scattering–induced transparency configuration, a tapered silica optical fiber is evanescently coupled to a high-$Q$ silica microsphere that supports two optical whispering-gallery modes $(\omega_1,k_1)$ and $(\omega_2,k_2)$, together with an acoustic resonance $(\Omega,q)$. Phase matching requires $\omega_2 \to \omega_1 + \Omega$ and $k_2 \to k_1 + q$. In the forward direction these conditions select a single traveling-wave acoustic phonon, whereas in the backward direction no acoustic mode satisfies both energy and momentum conservation, so Brillouin coupling vanishes. The resulting Brillouin-scattering–induced transparency is therefore unidirectional, and the backward probe sees ordinary resonant absorption (Kim et al., 2016).

In traveling-wave electro-optic devices, nonreciprocity is produced by the different accumulated phase seen by light that co-propagates with a microwave and light that counter-propagates with it. A single long thin-film lithium niobate phase modulator driven by a single-tone microwave achieves isolation because the forward optical wave counter-propagates relative to the microwave and, under the right choice of $f_{\rm RF}$, the net RF-induced phase accumulation vanishes; the backward optical wave co-propagates with the microwave, accumulates a large phase modulation, and the carrier is depleted into sidebands when $J_0(\beta)=0$ with $\beta \approx 0.765\pi$ (Yu et al., 2022). In the four-arm dynamic rotating destructive interference architecture, the forward beam again integrates to zero net modulation, while the backward beam acquires a non-zero, time-varying phase that is canceled by phasor recombination among four arms driven with quarter-period-shifted RF signals (Han et al., 2 Sep 2025).

A closely related mechanism appears in spatiotemporal silicon and lithium-niobate devices based on traveling refractive-index modulation. Lira et al. formulate the perturbation as $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$, and only one propagation direction simultaneously satisfies energy and momentum conservation for an indirect interband photonic transition between even and odd modes (Lira et al., 2011). In the LNOI two-modulator scheme, the forward field is left unmodulated when $(\omega_1,k_1)$0, whereas the reverse field is expanded in Bessel sidebands and the carrier can be strongly suppressed before filtering (Huang et al., 2022).

Traveling-wave isolation is not restricted to electro-optic or acousto-optic modulation. In the transverse magneto-optical Kerr effect coupled-waveguide system, the gyrotropy shifts the propagation constant of each waveguide by $(\omega_1,k_1)$1, so the supermode decomposition is different for forward and backward propagation; at the calibrated condition $(\omega_1,k_1)$2 and $(\omega_1,k_1)$3, a $(\omega_1,k_1)$4 relative phase shift is produced between forward and backward coupling (Chao et al., 14 Feb 2025). In the nonlinear adiabatic scheme proposed by Rangelov and Longhi, a chirped quasi-phase-matching profile produces forward phase matching and adiabatic sum-frequency conversion into a strongly absorbed ultraviolet wave, while backward propagation does not encounter the same phase-matching trajectory (Rangelov et al., 2016).

2. Canonical device architectures

The traveling-wave optical isolator is therefore a family of structures rather than a single topology. The representative architectures reported in the literature differ mainly in whether nonreciprocity is generated by photon–phonon coupling, microwave-driven spatiotemporal modulation, gyrotropy, or nonlinear adiabatic conversion.

Mechanism	Representative structure	Distinctive condition
BSIT	tapered silica optical fiber coupled to a high-$(\omega_1,k_1)$5 silica microsphere	$(\omega_1,k_1)$6, $(\omega_1,k_1)$7
EO traveling-wave depletion	single long TFLN phase modulator with CPW electrode	$(\omega_1,k_1)$8
DRDI	four-arm Mach–Zehnder interferometer in Si$(\omega_1,k_1)$9N$(\omega_2,k_2)$0 with TFLN	$(\omega_2,k_2)$1
LNOI spatiotemporal isolator	two cascaded travelling wave phase modulators plus ring resonator	$(\omega_2,k_2)$2
EMP acoustic isolator	triply-guided suspended silicon beam	$(\omega_2,k_2)$3
TMOKE modal-beating isolator	evanescent-coupled silicon waveguides with MO garnet	$(\omega_2,k_2)$4, $(\omega_2,k_2)$5
Nonlinear adiabatic isolator	chirped QPM waveguide with strong absorption at $(\omega_2,k_2)$6	$(\omega_2,k_2)$7 swept through resonance
Unidirectional resonant tunneling	ENG$(\omega_2,k_2)$8$(\omega_2,k_2)$9MO%%%%3$Q$3%%%%1ENG$(\Omega,q)$2	$(\Omega,q)$3, $(\Omega,q)$4

The Brillouin-scattering–induced transparency isolator is a waveguide–resonator microsystem in which the footprint is set by a microsphere of diameter $(\Omega,q)$5–$(\Omega,q)$6 and a tapered waveguide; its operation is explicitly material-agnostic and wavelength-agnostic because it relies only on the generic photoelastic interaction in dielectric media (Kim et al., 2016). The single-modulator TFLN device is structurally simpler: it consists of a single long traveling-wave phase modulator, a coplanar-waveguide electrode, and mode converters for low-loss coupling (Yu et al., 2022). The DRDI device instead uses four parallel optical channels, two push-pull modulator pairs, buried coplanar RF electrodes, and six on-chip chromium microheaters to tune amplitudes and static phases (Han et al., 2 Sep 2025).

Other implementations require explicit modal filtering. The silicon indirect-transition isolator uses a slotted waveguide supporting even and odd modes, followed by filtering so that phase-matched backward conversion becomes attenuation at the system level (Lira et al., 2011). The EMP acoustic isolator uses a triply-guided suspended silicon beam carrying two optical modes and one horizontal-shear acoustic mode at $(\Omega,q)$7, with adiabatic tapers or two-mode multiplexers and demultiplexers at the ends of the active section (Dostart et al., 2018). The LNOI spatiotemporal isolator similarly requires an add-drop ring resonator so that reverse-propagating sidebands are dropped while the forward carrier passes (Huang et al., 2022).

Magneto-optic traveling-wave isolators remain a distinct branch. Moccia, Castaldi, Galdi and Engheta analyze an ENG–MO–ENG tri-layer in the Faraday configuration, where the ENG layers act as high-reflectivity mirrors and the MO layer provides different propagation constants for right- and left-circularly-polarized eigenmodes; the tiny Faraday-induced phase shift is amplified by cavity round trips (Moccia et al., 2013). The TMOKE coupled-waveguide proposal instead avoids resonators and interferometers, using modal beating in evanescent-coupled silicon waveguides with one garnet-clad wall per guide (Chao et al., 14 Feb 2025).

3. Theoretical descriptions and figures of merit

Most traveling-wave optical isolators are analyzed with coupled-mode equations, transfer matrices, or harmonic expansions that make the forward/backward asymmetry explicit. In the Brillouin-scattering–induced transparency model, the intracavity optical amplitudes $(\Omega,q)$8 and traveling phonon amplitude $(\Omega,q)$9 satisfy coupled equations, and the pump-enhanced coupling rate is defined as $\omega_2 \to \omega_1 + \Omega$0. The normalized probe transmission is

$\omega_2 \to \omega_1 + \Omega$1

Forward transmission uses $\omega_2 \to \omega_1 + \Omega$2, whereas backward transmission uses $\omega_2 \to \omega_1 + \Omega$3. The isolation ratio and insertion loss are then

$\omega_2 \to \omega_1 + \Omega$4

At $\omega_2 \to \omega_1 + \Omega$5, the transparency window opens because the term $\omega_2 \to \omega_1 + \Omega$6 cancels the bare optical loss (Kim et al., 2016).

In traveling-wave EO depletion, the phase accumulated by the optical field is written as an integral of the microwave drive along the electrode. When $\omega_2 \to \omega_1 + \Omega$7 and microwave loss is negligible, the forward phase integrates to zero. The backward phase modulation depth is $\omega_2 \to \omega_1 + \Omega$8, and setting $\omega_2 \to \omega_1 + \Omega$9 produces $k_2 \to k_1 + q$0, so the carrier is completely depleted into higher-order sidebands (Yu et al., 2022). The LNOI two-modulator formulation writes the backward field as a Bessel-series expansion,

$k_2 \to k_1 + q$1

which directly connects isolation to reverse-carrier suppression before ring filtering (Huang et al., 2022).

The DRDI formalism expresses the forward transmittance as

$k_2 \to k_1 + q$2

while the backward transmittance is

$k_2 \to k_1 + q$3

By construction, $k_2 \to k_1 + q$4 ideally at all $k_2 \to k_1 + q$5, and the time-averaged isolation is

$k_2 \to k_1 + q$6

This description emphasizes a distinction within the traveling-wave category: some devices isolate by carrier depletion and filtering, whereas others isolate by direct destructive interference of the reverse field (Han et al., 2 Sep 2025).

Acoustic and modal-conversion isolators use analogous coupled-wave models. In the EMP isolator, two optical mode amplitudes $k_2 \to k_1 + q$7 and $k_2 \to k_1 + q$8 are coupled by an electrically driven acoustic amplitude $k_2 \to k_1 + q$9, with phase matching set by $f_{\rm RF}$0 (Dostart et al., 2018). In the silicon indirect-transition device, even and odd optical modes satisfy coupled equations with $f_{\rm RF}$1 and $f_{\rm RF}$2, and exact conversion requires $f_{\rm RF}$3 and $f_{\rm RF}$4 (Lira et al., 2011). In the TMOKE coupled-waveguide system, the forward and backward transfer matrices differ because $f_{\rm RF}$5, even though the coupling coefficient $f_{\rm RF}$6 is unchanged (Chao et al., 14 Feb 2025).

4. Experimental and simulated performance landscape

The reported figures of merit span microscale resonant devices, centimeter-scale traveling-wave modulators, and broadband spatiotemporal architectures. The comparison below reproduces representative results exactly as reported.

Work	Reported result	Status
BSIT microscale isolator (Kim et al., 2016)	forward loss: $f_{\rm RF}$7 dB; isolation contrast: $f_{\rm RF}$8 dB over $f_{\rm RF}$9 kHz bandwidth; footprint: $J_0(\beta)=0$0 sphere + tapered waveguide	experiment
TFLN EO isolator (Yu et al., 2022)	maximum optical isolation of $J_0(\beta)=0$1 dB and an on-chip insertion loss of $J_0(\beta)=0$2 dB; isolation ratio larger than $J_0(\beta)=0$3 dB across $J_0(\beta)=0$4 to $J_0(\beta)=0$5 nm	experiment
DRDI isolator (Han et al., 2 Sep 2025)	about $J_0(\beta)=0$6 dB isolation at $J_0(\beta)=0$7 nm; over $J_0(\beta)=0$8 dB isolation across an approximately $J_0(\beta)=0$9 nm bandwidth; $\beta \approx 0.765\pi$0 dB isolation for two simultaneous lasers within an approximately $\beta \approx 0.765\pi$1 nm wavelength window	experiment
LNOI spatiotemporal isolator (Huang et al., 2022)	isolation of $\beta \approx 0.765\pi$2 dB	experiment
Silicon indirect-transition isolator (Lira et al., 2011)	contrast ratio exceeds $\beta \approx 0.765\pi$3 dB in simulations; experimentally observe a strong contrast up to $\beta \approx 0.765\pi$4 dB	simulation + experiment
EMP acoustic isolator (Dostart et al., 2018)	over $\beta \approx 0.765\pi$5 dB of isolation and $\beta \approx 0.765\pi$6 dB of insertion loss with $\beta \approx 0.765\pi$7 GHz optical bandwidth and a $\beta \approx 0.765\pi$8 cm device length; $\beta \approx 0.765\pi$9 mW of electrical drive power	theory/simulation
TMOKE coupled-waveguide isolator (Chao et al., 14 Feb 2025)	length of $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$0; $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$1 dB-isolation bandwidth as high as $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$2 nm; peak isolation $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$3 dB at $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$4 nm; insertion loss $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$5–$\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$6 dB	simulation
Nonlinear adiabatic isolator (Rangelov et al., 2016)	peak isolation up to $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$7 dB; isolation $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$8 dB over $\Delta n(z,t)=\Delta n_0 \cos(\Omega t-qz)$9 nm around $(\omega_1,k_1)$00 nm in a $(\omega_1,k_1)$01 cm device; insertion loss negligible $(\omega_1,k_1)$02 dB, backward-pass)	theory
ENG–MO–ENG CP isolator (Moccia et al., 2013)	$(\omega_1,k_1)$03 dB; $(\omega_1,k_1)$04 dB; $(\omega_1,k_1)$05 nm	design example

These data show that the term “traveling-wave optical isolator” covers markedly different operating regimes. Resonant Brillouin devices achieve ultralow forward loss in a microscale footprint but operate over sub-megahertz windows (Kim et al., 2016). Single-modulator TFLN EO devices achieve both very low on-chip insertion loss and broad C+L-band isolation (Yu et al., 2022). DRDI sacrifices the use of resonant optical elements in favor of an approximately $(\omega_1,k_1)$06 nm operating window in the visible, including simultaneous two-laser isolation (Han et al., 2 Sep 2025). By contrast, the LNOI two-modulator-plus-ring device is explicitly narrowband because the optical isolation bandwidth is set by the ring resonance (Huang et al., 2022).

5. Design trade-offs and integration constraints

Design optimization in traveling-wave optical isolation is dominated by a small set of recurring trade-offs: phase matching versus bandwidth, coupling strength versus drive power, insertion loss versus footprint, and nonreciprocity versus fabrication tolerance. In the microsphere BSIT device, resonator $(\omega_1,k_1)$07 sets the isolation bandwidth because $(\omega_1,k_1)$08 should be chosen so that $(\omega_1,k_1)$09 lies in the target isolation bandwidth; lower $(\omega_1,k_1)$10 gives broader bandwidth but requires larger $(\omega_1,k_1)$11 to open a full transparency. The phonon $(\omega_1,k_1)$12 controls the pump threshold for transparency, and the waveguide–resonator gap must be adjusted to achieve $(\omega_1,k_1)$13 for critical coupling and maximum backward extinction. Increasing $(\omega_1,k_1)$14 broadens the transparency window from $(\omega_1,k_1)$15 toward $(\omega_1,k_1)$16, reduces forward loss, and requires pump-power management to avoid thermal bistability (Kim et al., 2016).

In long EO traveling-wave modulators, velocity matching is decisive. The TFLN isolator engineers $(\omega_1,k_1)$17 to maintain microwave–optical overlap over lengths of $(\omega_1,k_1)$18 to $(\omega_1,k_1)$19 cm, and its reported optical bandwidth follows from the residual mismatch between RF and optical group velocities (Yu et al., 2022). The LNOI spatiotemporal device likewise relies on $(\omega_1,k_1)$20 tuned to match $(\omega_1,k_1)$21, but because it uses a ring resonator for filtering, the system bandwidth is ultimately narrowed by the resonant filter even if the modulators themselves have RF bandwidth $(\omega_1,k_1)$22 GHz (Huang et al., 2022). This distinction suggests that traveling-wave phase modulation alone does not guarantee broadband optical isolation; the overall bandwidth is set by the narrowest subsystem.

Broadband nonresonant EO schemes shift complexity from optical resonators to RF synthesis and thermal trimming. In the DRDI device, the bandwidth is set only by the wavelength dependence of the passive $(\omega_1,k_1)$23 couplers, any dispersion in the RF line, and the extent to which the $(\omega_1,k_1)$24 condition holds versus $(\omega_1,k_1)$25; over $(\omega_1,k_1)$26–$(\omega_1,k_1)$27 nm, $(\omega_1,k_1)$28 dB isolation is maintained after minor heater retuning at each wavelength (Han et al., 2 Sep 2025). The EMP acoustic isolator has a different constraint: acoustic attenuation $(\omega_1,k_1)$29 limits propagation to $(\omega_1,k_1)$30 mm, so ten piezoelectric transducers are placed in series, each re-injecting the acoustic tone with the correct RF phase (Dostart et al., 2018).

Magneto-optic and resonant-tunneling devices have their own fabrication sensitivities. In the TMOKE coupled-waveguide system, smooth sidewall coverage and tight control of gap $(\omega_1,k_1)$31 $(\omega_1,k_1)$32 are crucial to set $(\omega_1,k_1)$33, and the design explicitly trades off loss and footprint by balancing waveguide width against evanescent overlap with the MO garnet (Chao et al., 14 Feb 2025). In the ENG–MO–ENG cavity, bandwidth can be broadened by using thinner ENG or lower $(\omega_1,k_1)$34, but this comes at the expense of isolation; rounding $(\omega_1,k_1)$35 to $(\omega_1,k_1)$36 alters $(\omega_1,k_1)$37 by $(\omega_1,k_1)$38 and isolation by $(\omega_1,k_1)$39, whereas $(\omega_1,k_1)$40 begins to degrade insertion loss by $(\omega_1,k_1)$41 dB (Moccia et al., 2013).

The nonlinear adiabatic isolator presents a different optimization criterion: the chirp rate must satisfy the adiabatic condition $(\omega_1,k_1)$42 at resonance, and the optimum balance between coupling and absorption is achieved when $(\omega_1,k_1)$43 so that the adiabatic transfer is neither under-damped nor over-damped (Rangelov et al., 2016). This suggests that, across traveling-wave architectures, the key engineering variable is not merely “nonreciprocity strength,” but the controlled co-design of propagation, coupling, and dissipation.

6. Conceptual distinctions, limitations, and applications

Several recurring misconceptions are clarified by the reported literature. First, traveling-wave optical isolation is not synonymous with magnet-free isolation. The BSIT, TFLN, DRDI, LNOI spatiotemporal, silicon indirect-transition, EMP acoustic, and nonlinear adiabatic schemes are explicitly magnet-free (Kim et al., 2016, Yu et al., 2022, Han et al., 2 Sep 2025, Lira et al., 2011, Dostart et al., 2018, Rangelov et al., 2016), but the ENG–MO–ENG resonant-tunneling design and the TMOKE coupled-waveguide design rely on magneto-optical materials and biasing (Moccia et al., 2013, Chao et al., 14 Feb 2025). Second, traveling-wave operation is not necessarily broadband. The LNOI two-modulator isolator is narrowband because the optical isolation bandwidth is set by the ring resonance $(\omega_1,k_1)$44 MHz (Huang et al., 2022), while the microsphere BSIT device reports $(\omega_1,k_1)$45 kHz isolation bandwidth in its strong-pump case (Kim et al., 2016). By contrast, the TFLN EO isolator maintains $(\omega_1,k_1)$46 dB isolation over $(\omega_1,k_1)$47–$(\omega_1,k_1)$48 nm and the DRDI device maintains over $(\omega_1,k_1)$49 dB isolation across an approximately $(\omega_1,k_1)$50 nm bandwidth (Yu et al., 2022, Han et al., 2 Sep 2025).

A further distinction concerns linearity. The BSIT microsphere device explicitly demonstrates complete linear optical isolation (Kim et al., 2016), and the silicon indirect-transition isolator is reported as linear with respect to signal light, with the observed contrast ratio independent of the timing, the format, the amplitude and the phase of the input signal (Lira et al., 2011). The nonlinear adiabatic scheme, by contrast, uses undepleted-pump three-wave mixing and strong absorption at the generated sum-frequency wave; Rangelov and Longhi explicitly frame it as avoiding the limitations of dynamic reciprocity found in other nonlinear optical isolation methods (Rangelov et al., 2016). This suggests that “traveling-wave isolator” denotes a propagation geometry rather than a unique reciprocity-breaking mechanism.

The application space is broad but technically specific. The central use is protection against back-reflections and optical feedback in coherent laser systems and photonic integrated circuits (Yu et al., 2022, Han et al., 2 Sep 2025). The TFLN EO device verifies that a hybrid DFB laser–LN isolator module protects the single-mode operation and the linewidth of the DFB laser from reflection (Yu et al., 2022). DRDI is positioned for atomic spectroscopy, laser cooling, and locking applications because the $(\omega_1,k_1)$51–$(\omega_1,k_1)$52 nm span covers key alkali atomic transitions and supports simultaneous isolation for two lasers (Han et al., 2 Sep 2025). The LNOI spatiotemporal device is presented as suitable for integration with III–V laser diodes and Erbium doped gain sections (Huang et al., 2022). The EMP, TFLN, and TMOKE platforms are all framed as candidates for coherent communications, microwave photonics, sensing, and large-scale photonic processors (Dostart et al., 2018, Yu et al., 2022, Chao et al., 14 Feb 2025).

Taken together, the literature indicates that traveling-wave optical isolation has become a unifying design paradigm for nonreciprocal photonics rather than a single device category. The unifying principle is directional asymmetry in an extended interaction region; the differentiating factors are whether that asymmetry is generated by Brillouin scattering, electro-optic spatiotemporal modulation, electrically driven acoustic waves, gyrotropy, resonant photon tunneling, or nonlinear adiabatic conversion, and whether the system-level priority is ultralow loss, broad bandwidth, compact footprint, low electrical power, or compatibility with monolithic photonic integration.