X-Cut TFLN: Anisotropic Photonics Platform

• X-cut TFLN is a thin-film lithium niobate platform defined by a unique x-axis cut that imparts anisotropic optical properties.
• It enables precise control over modal polarization and orientation-dependent tuning critical for electro-optic, acousto-optic, and nonlinear applications.
• Advances in fabrication and domain engineering have minimized losses, yielding high-Q resonators and efficient nonlinear optical interactions.

X-cut thin-film lithium niobate (TFLN) is a thin-film lithium-niobate platform in which the wafer surface normal is along the crystal $x$-axis, so the film plane contains the $y$- and $z$-axes. In this geometry, propagation direction and modal polarization are not secondary layout choices but primary device variables, because the guided field samples the ordinary and extraordinary axes differently as the circuit rotates through the wafer plane. X-cut TFLN is accordingly treated in the recent literature as a cut-specific integrated-photonics platform rather than a generic subset of lithium-niobate-on-insulator, with demonstrated roles in electro-optic isolation and modulation, anisotropic thermo-optic control, acousto-optics, periodic-poling-based nonlinear optics, soliton microcombs, and deterministic rare-earth activation (Shim et al., 1 Jan 2026, Gao et al., 2023, Blight et al., 6 Nov 2025).

1. Crystallographic basis and anisotropic modal physics

In x-cut lithium niobate, the crystal $x$-axis is normal to the wafer surface, while the $y$- and $z$-axes lie in the film plane. The $x$- and $y$-axes are associated with the ordinary refractive index $n_o$, and the $z$-axis is associated with the extraordinary refractive index $y$0. For x-cut TFLN waveguides, propagation is therefore considered in the $y$1-$y$2 plane, often parameterized by an angle $y$3 measured with respect to the crystal $y$4-axis. This makes the effective optical response orientation dependent even for nominally identical waveguide cross-sections (Shim et al., 1 Jan 2026, Ruesing et al., 2019).

The most compact analytical statement of this anisotropy appears in the generalized thermo-optic model for x-cut TFLN, where the effective-index tuning is written as

$y$5

Under the model’s angle-invariant local-field approximation,

$y$6

Because $y$7, TE-like modes in x-cut TFLN can vary strongly with $y$8, whereas TM-like modes remain more weakly angle dependent because they are mostly ordinary-like (Shim et al., 1 Jan 2026).

This anisotropy is not confined to thermal tuning. In periodically poled x-cut films, the in-plane $y$9-axis coincides with the optical axis / polar axis, and a field polarized along in-plane $z$0 directly addresses $z$1. That is the explicit reason x-cut is described as attractive for integrated nonlinear optics and why TE-like guided fields in x-cut waveguides are repeatedly used to access the strongest $z$2 interaction (Ruesing et al., 2019, Doshi et al., 11 Apr 2025). A related complication appears in curved resonators: for TE modes in x-cut microrings, the effective $z$3 interaction varies around the ring because the local propagation direction rotates relative to the crystal axes, and the supplementary analysis of the material-loss study notes that TE/TM avoided crossings and adiabatic conversion can occur for propagation angles away from the crystalline axes (Shams-Ansari et al., 2022). This suggests that x-cut TFLN is best understood as an orientation-defined design space rather than a single isotropic waveguide platform.

2. Wafer platforms, fabrication routes, and optical-loss limits

The x-cut TFLN literature spans several layer stacks. Representative examples include $z$4 nm x-cut LN on $z$5m thermal SiO$z$6 on Si for ultra-low-loss microrings, $z$7 nm x-cut TFLN on $z$8m buried oxide for electro-optic isolation and modulator arrays, $z$9 nm x-cut TFLN on $x$0m buried oxide for acousto-optics, $x$1 nm x-cut MgO-doped TFLN on $x$2m buried oxide for nanodomain poling, and $x$3 nm ultra-thin x-cut LN on high-resistivity silicon for NEMS resonators (Shams-Ansari et al., 2022, Gao et al., 2023, Ni et al., 3 Jun 2026, Sabatti et al., 17 Jul 2025, Tetro et al., 2024).

A central quantitative benchmark for x-cut TFLN is the separation between measured cavity loss and inferred material loss. In half-etched $x$4 nm x-cut microrings with radius $x$5m, width $x$6m, and target etch depth $x$7 nm, three process variants yielded mean intrinsic TE-mode quality factors of $x$8, $x$9, and $y$0. Using Kerr-calibrated photothermal measurements, the same study extracted $y$1 MHz, $y$2 MHz, and $y$3 MHz for the three samples, corresponding in the best case to a material-limited quality factor $y$4 and propagation loss $y$5. The paper is explicit that the measured resonators remain limited mainly by line-edge roughness-induced scattering and Rayleigh back-scattering rather than by an intrinsic x-cut material floor (Shams-Ansari et al., 2022).

Process engineering in x-cut TFLN has correspondingly focused on post-etch damage reduction and surface smoothing. The loss-reduction work recommends annealing in O$y$6 at $y$7C for $y$8 h and preserving the benefit with low-temperature $y$9C ICPCVD SiO$z$0 cladding followed by re-annealing (Shams-Ansari et al., 2022). A complementary route is isotropic atomic layer etching of x-cut MgO-doped LN using sequential H$z$1 and SF$z$2/Ar plasmas. That process reports an etch rate of $z$3 nm/cycle with a synergy of $z$4, and when applied as a post-processing treatment to TFLN waveguides etched by Ar$z$5 milling it reduces sidewall RMS roughness from $z$6 nm to $z$7 nm, with a lateral etch rate of $z$8 nm/cycle on each side (Chen et al., 2023). The same paper also shows the tradeoff: smoothing is accompanied by isotropic feature erosion and increased roughness on initially flat bulk surfaces. This suggests that the principal fabrication question in x-cut TFLN is no longer whether low loss is fundamentally possible, but how closely practical processes can approach the material limit without sacrificing dimensional control.

3. Electro-optic, thermo-optic, and acousto-optic control

X-cut TFLN is repeatedly chosen when the objective is strong field-driven functionality on chip. A direct example is the electro-optic isolator fabricated on a $z$9-nm-thick x-cut TFLN bonded to a $x$0m SiO$x$1 layer, using a $x$2 mm ground-signal-ground coplanar traveling-wave electrode and a $x$3m-top-width LN ridge waveguide. The device demonstrates $x$4 dB isolation at $x$5 GHz and $x$6 dBm, remains above $x$7 dB isolation from $x$8 nm to $x$9 nm and from $y$0 GHz to $y$1 GHz, and shows $y$2 dB fiber-to-fiber insertion loss (Gao et al., 2023). The underlying nonreciprocal modulation is described through

$y$3

with the carrier-suppression condition $y$4 at $y$5, and an isolation relation

$y$6

Within that work, x-cut is not treated as a passive substrate choice but as the enabling electro-optic medium.

A more system-level example is the hybrid-integrated $y$7 x-cut TFLN modulator array. Fabricated on a commercial x-cut lithium-niobate-on-insulator wafer with a $y$8 nm LN thin film bonded to a buried SiO$y$9 layer on a $n_o$0m Si substrate, it combines a three-stage cascaded $n_o$1 multimode-interference splitter, eight $n_o$2 mm traveling-wave Mach-Zehnder modulators, thermal tuning electrodes, on-chip $n_o$3 terminations, and passive butt-coupling to a $n_o$4 nm DFB laser. The splitter shows a maximum normalized power deviation of $n_o$5, all channels exceed $n_o$6 GHz electro-optic $n_o$7 dB bandwidth, the measured half-wave voltages are $n_o$8–$n_o$9 V for $z$0–$z$1 V cm, and the extinction ratio reaches approximately $z$2 dB. The bare-chip insertion loss is $z$3–$z$4 dB, while DFB bonding adds approximately $z$5 dB coupling loss (Hu et al., 20 May 2026).

The same orientation dependence appears in slower control and in phononic transduction. The anisotropic thermo-optic model for x-cut TFLN was validated experimentally on heater-integrated racetrack resonators and fit with $z$6, confirming that TE tuning is maximal near $z$7 and minimal near $z$8, while TM tuning is weaker and less angle sensitive (Shim et al., 1 Jan 2026). In acousto-optics, a non-suspended push-pull acousto-optic modulator on a $z$9 nm x-cut film was fabricated with devices at $y$00, $y$01, and $y$02, where $y$03 is the in-plane acoustic propagation angle relative to the crystal $y$04-axis. The optimized $y$05 device achieved $y$06 at $y$07 GHz with a $y$08m interaction length and a $y$09 MHz bandwidth, directly linking electromechanical performance to x-cut orientation engineering (Ni et al., 3 Jun 2026). Together these results show that x-cut TFLN is defined as much by direction-sensitive control laws as by its layer stack.

4. Domain engineering and $y$10 nonlinear optics

Periodic poling is a central mechanism by which x-cut TFLN accesses strong second-order interactions. For near-IR to visible SHG, a recent SHM-based metrology study used a $y$11 nm-thick x-cut MgO-doped TFLN layer on $y$12m buried oxide, with waveguides along the crystal $y$13-axis and TE$y$14 polarized along the crystal $y$15-axis so that the $y$16 tensor element is used. In that platform, efficient QPM required a poling period $y$17m with period variation below $y$18 nm. The paper shows that increasing the SHM raster scan step size from $y$19 nm to $y$20 nm gives a $y$21 imaging speedup, and that sampling only $y$22 fields of $y$23m each—about $y$24 periods, or $y$25 of a $y$26 mm grating—is sufficient to predict SH output accurately. The device reached a measured conversion efficiency of $y$27, compared with a corrected prediction of $y$28 (Doshi et al., 11 Apr 2025).

The optical interpretation of SH microscopy in x-cut TFLN is itself cut specific. In a $y$29 nm x-cut LN film on $y$30m SiO$y$31 on silicon, the dominant back-reflected SH signal was shown not to be the directly generated counter-propagating SH field familiar from bulk ferroelectric microscopy, but the much stronger co-propagating SH field redirected by buried-interface reflections. At an $y$32 nm pump, the paper reports $y$33 nm and $y$34m, together with $y$35. This enables SH microscopy in x-cut TFLN to distinguish complete from partial inversion with depth sensitivity down to tens of nanometers (Ruesing et al., 2019). That result is directly relevant to submicron periodic poling, where inversion depth rather than lateral period alone becomes the limiting variable.

Two recent nonlinear-device papers push this logic further. A continuous-wave-pumped optical parametric amplifier on a $y$36-inch, $y$37 nm x-cut TFLN wafer uses domain engineering with $y$38m in a $y$39 mm waveguide to phase match both on-chip SHG pump generation and OPA in TE$y$40. It reports $y$41 dB on-chip gain, $y$42 dB net gain, and more than $y$43 nm $y$44-dB bandwidth across $y$45–$y$46 nm, with insertion loss $y$47 dB and propagation loss on the order of $y$48 dB/cm (Chen et al., 2024). At the opposite extreme of QPM period, sidewall poling in x-cut TFLN enabled scalable periods of $y$49 nm and $y$50 nm. Using a $y$51 nm-thick, $y$52 MgO-doped x-cut film on $y$53m SiO$y$54, the authors fabricated a $y$55 nm counter-propagating device with near-$y$56 duty cycle and a $y$57 nm backward-propagating device with partial-depth inversion. They report normalized conversion efficiencies of $y$58 and $y$59, respectively, and SPDC brightnesses of $y$60 and $y$61, with coincidence-to-accidental ratio around $y$62 at $y$63 mW (Sabatti et al., 17 Jul 2025).

Backward-wave nonlinear optics has also been demonstrated in an $y$64 nm-thick x-cut TFLN platform with periodic poling before etching. There, a physical period of $y$65 nm over a $y$66 mm interaction length realizes third-order QPM, corresponding to an effective first-order period of $y$67 nm. The same waveguide supports backward-wave SHG near $y$68 nm and backward-wave DFG with a pump near $y$69 nm, a counter-propagating signal near $y$70 nm, and idler generation from $y$71 nm to $y$72 nm. The paper reports $y$73 conversion efficiency for backward SHG, $y$74 for BWDFG, and a temperature tuning slope $y$75 (Koyaz et al., 14 Jun 2026). These results collectively indicate that x-cut TFLN now spans ordinary forward-QPM devices, CW-pumped OPA, and first-order or higher-order counter-propagating and backward-wave interactions.

5. Resonators, surface-wave optics, soliton combs, and high-frequency electromechanics

X-cut TFLN is also a resonant and wave-based platform beyond straight $y$76 circuits. In Bloch-surface-wave optics, a one-dimensional photonic crystal with four pairs of Si$y$77N$y$78 ($y$79 nm) and SiO$y$80 ($y$81 nm) plus a $y$82 nm x-cut TFLN top layer sustains TE-polarized BSWs at $y$83 nm. The theoretical and experimental coupling angle is $y$84, and propagation was detected over a distance of $y$85 mm. The same paper explicitly links x-cut orientation to future electro-optic use by noting $y$86 pm/V at $y$87 nm and stating that for x-cut TFLN the electrodes should be placed along the $y$88 axis, so BSWs should propagate along the $y$89 direction (Kovalevich et al., 2018). This is not yet a full electro-optic device, but it establishes x-cut TFLN as an active top layer for telecom-band dielectric surface waves.

In Kerr-resonator physics, x-cut TFLN had long been considered difficult because of multiple strong Raman-active modes, in-plane refractive-index anisotropy, and photorefractive effects. Stable soliton microcombs on x-cut lithium niobate addressed this by using racetrack resonators whose straight sections are aligned along the crystal $y$90-axis so that the TE$y$91 field aligns predominantly with the ordinary $y$92-axis, reducing the peak gain of the $y$93 Raman mode by more than $y$94 dB compared with alignment to the extraordinary $y$95-axis. With a $y$96 nm starting film stack, a main-pump TE$y$97 resonance at $y$98 nm, and a counter-propagating TM$y$99 cooler pump at $z$00 nm, the work demonstrated single-soliton and crystal states at $z$01 GHz, $z$02 GHz, $z$03 GHz, $z$04 GHz, $z$05 GHz, and $z$06 GHz. The main TE$z$07 mode had intrinsic $z$08, the measured resonance linewidth was about $z$09 MHz, the coupling gap was $z$10m, and the sidewall angle about $z$11 (Song et al., 18 Feb 2025). This is a cut-specific milestone because the same paper argues that efficient EO devices are substantially lacking on the Z-cut platform compared with X-cut.

At still shorter wavelengths in the acoustic domain, ultra-thin x-cut LN supports multifrequency NEMS resonators. Laterally vibrating resonators and degenerate LVRs fabricated on a $z$12 nm x-cut film on high-resistivity silicon, with propagation along YZ30° for $z$13 modes and YZ-10° for $z$14 modes, cover $z$15 to $z$16 GHz on a single chip. The devices use acoustic wavelengths from $z$17 to $z$18 nm and achieve $z$19 up to $z$20, $z$21 up to $z$22, electromechanical coupling $z$23 up to $z$24, and figure of merit $z$25 up to $z$26, with $z$27 across roughly $z$28–$z$29 GHz (Tetro et al., 2024). A plausible implication is that x-cut TFLN is now being used coherently across optical, electro-optic, and electromechanical frequency scales rather than within a single subfield.

6. Active-ion integration, low-temperature response, and limits of transfer from non-X-cut literature

X-cut TFLN is increasingly being treated as an active material platform rather than only a passive or field-driven one. A recent study of ion-implanted erbium in x-cut TFLN used focused ion beam implantation of isotopically selected $z$30 at $z$31 kV and $z$32 keV into congruent undoped x-cut LNOI, with doses from $z$33 to $z$34. SRIM gave a mean implantation depth of $z$35 nm, annealing at $z$36C for $z$37 min in $z$38 sccm oxygen activated the emitters without observable feature broadening, and photoluminescence from implanted regions showed Stark-split $z$39-$z$40 transitions comparable with bulk Er:LiNbO$z$41. The integrated PL intensity was proportional to dose, while temperature-dependent PL from $z$42 K to $z$43 K showed conventional behavior down to approximately $z$44 K, followed by a marked decrease in emission intensity and lifetime. The authors attribute this anomaly to suppression of the pyroelectric response in LiNbO$z$45 at low temperatures, which affects local electric fields and therefore Er$z$46 emission (Blight et al., 6 Nov 2025). This directly links rare-earth spectroscopy in x-cut TFLN to the same internal-field physics that already governs its EO and thermo-optic behavior.

At the same time, a recurrent source of confusion in the TFLN literature is the assumption that all thin-film LN papers can be read as x-cut papers. Several prominent works in adjacent areas explicitly cannot. The photonic-crystal nanocavity laser on erbium-doped TFLN uses a Z-cut wafer and does not discuss x-cut-specific issues such as in-plane optical-axis orientation or TE/TM mapping to ordinary and extraordinary indices (Liu et al., 2023). The TFLN AWG fabricated by PLACE is explicitly on a Z-cut platform, and its design rationale relies on Z-cut TE isotropy in the wafer plane (Wang et al., 2023). The monolithically integrated ultra-high-$z$47 microring by PLACE is likewise fabricated on a $z$48 nm-thick Z-cut TFLN wafer (Li et al., 2023), and the earlier optomechanical disk-resonator work starts from a Z-cut LiNbO$z$49 wafer (Wang et al., 2014). Even the active–passive tiled TFLN amplifier array does not specify whether the films are x-cut, y-cut, or z-cut (Zhou et al., 2022). These papers remain informative for fabrication flow, passive-device architecture, or generic TFLN device physics, but they do not by themselves establish x-cut design rules.

This distinction matters because x-cut behavior is often orientation specific. TE/TM assignment, access to $z$50 or $z$51, anisotropic thermo-optic tuning, acoustic propagation angle, ring-mode evolution, and Raman response are all explicitly shown in the x-cut literature to depend on propagation direction or local field projection onto the crystal axes (Shim et al., 1 Jan 2026, Ruesing et al., 2019, Ni et al., 3 Jun 2026, Song et al., 18 Feb 2025). The most reliable current picture is therefore that x-cut TFLN is not defined only by using lithium niobate in thin-film form; it is defined by a particular tensor geometry that must be carried through device layout, poling strategy, thermal control, and interpretation of measurements.