Low-Loss AlN Waveguide Fabrication

Updated 28 April 2026

The paper demonstrates that precise control over film quality, sidewall roughness, and defect passivation in AlN waveguides reduces propagation losses to as low as 0.12 dB/cm in the UV and achieves intrinsic Q-factors approaching one million.
It employs advanced epitaxial growth, hybrid buffer engineering, and high-resolution e-beam lithography combined with optimized etch protocols to minimize scattering and absorption losses.
Optimized waveguide geometry and dispersion engineering enable efficient mode confinement, supporting diverse applications like quantum optics, frequency conversion, and on-chip nonlinear photonics.

Low-loss aluminum nitride (AlN) waveguide fabrication constitutes the backbone of high-Q integrated photonics across ultraviolet (UV), visible, and telecom spectra. AlN offers a unique combination of ultra-wide transparency (bandgap ~6.1 eV), high refractive index, intrinsic second- (χ²) and third-order (χ³) optical nonlinearities, and mature process compatibility with both silicon (Si) and sapphire substrates. The reduction of propagation loss in AlN waveguides is essential for quantum optics, frequency conversion, UV spectroscopy, and on-chip nonlinear photonics. Achieving minimal loss requires atomic-level control over film quality, sidewall roughness, defect passivation, and cross-sectional geometry. Leading processes achieve propagation losses below 0.15 dB/cm at telecom, ~0.12 dB/cm in the UV, and 2 dB/cm in the visible, with intrinsic microring Q-factors approaching or exceeding one million.

1. Material Growth, Substrate Engineering, and Film Quality

The starting point for low-loss AlN waveguide fabrication is epitaxial or highly c-axis-textured AlN growth. High-quality AlN films are deposited on sapphire or silicon substrates with buffer engineering to mitigate lattice mismatch and suppress dislocation/void densities.

MOCVD on sapphire: Single-crystalline AlN films are grown on 2" c-plane (0001) sapphire with a 5–10 nm low-temperature AlN nucleation buffer (600°C), followed by high-temperature AlN (growth at 1150°C, 100 Torr, TMA/NH₃, V/III ~500). Resulting films exhibit <0.1° miscut, FWHM <300 arcsec, and dislocation density ∼10⁸ cm⁻² (Liu et al., 2018).
Hybrid buffer approaches: A 50 nm sputtered AlN buffer on sapphire enables subsequent MOVPE/MOCVD growth that is free of vertical voids, reducing scattering losses to <0.2 dB/cm (Brunetta et al., 7 Nov 2025).
Sputter deposition for Si platforms: High-purity, dual-cathode magnetron sputtering in an Ar/N₂ atmosphere (no external heating or <700°C) achieves c-axis orientation (rocking curve <2°), RMS surface roughness <2 nm, and stress tunable to ±75 MPa (Xiong et al., 2014, Singh et al., 2024).
Doped variants: Incorporation of Sc into AlN (Sc₀.₁Al₀.₉N) is achieved via sputter deposition on oxidized Si, with post-deposition CMP reducing roughness below 1 nm, improving Q_factors by ~2× (Wang et al., 2024).

Table 1. Representative AlN Film Growth Conditions and Properties

Growth Method	Substrate	Thickness	Crystallinity	Void/Defect Control
MOCVD	Sapphire (0001)	0.4–1.1 μm	Single crystal	Buffer/nucleation
Sputtering (AC/DC)	Si + SiO₂	0.2–0.65 μm	c-axis textured	RF stress control
Hybrid (sputter+MOCVD)	Sapphire (0001)	1.1 μm	Single crystal	Sputtered buffer layer

2. Lithography, Pattern Transfer, and Sidewall Engineering

The transition from planar AlN layers to photonic devices relies on high-resolution e-beam lithography and finely tuned plasma etching schemes.

Electron-beam lithography: ZEP-520A or HSQ resists enable sub-10 nm LER. 100 kV EBL with beam currents 2 nA and doses 500–2000 μC/cm² yields high-resolution, low-roughness patterns (Liu et al., 2018, Yan et al., 27 Jun 2025).
Charge mitigation: Conductive polymer layers (mr-Conductive, water-soluble) fully eliminate charging for AlN on sapphire, removed during develop, streamlining single-mask processes (Yan et al., 27 Jun 2025).
Mask Stack: Si₃N₄ or SiO₂ hard masks (100–250 nm) are deposited by low-stress sputtering for etch resistance and sidewall shape preservation.
ICP-RIE: Cl₂/BCl₃/Ar mixtures (ratios 10–25:6–20:5–9 sccm), power up to 800 W, bias ~–100 V, chamber pressure 3–5 mTorr, enable etch rates of 40–200 nm/min, selectivity AlN:SiO₂ or AlN:HSQ ≳ 2:1, and vertical (85–90°) sidewalls with <2 nm RMS roughness (Liu et al., 2018, Yan et al., 27 Jun 2025).
ALD/RTA treatments: ALD of 1 nm Al₂O₃ (10 cycles) and 400°C rapid thermal anneal drive loss from ~7 dB/cm to 2 dB/cm at 852 nm, without significant modification of mode confinement (Videnov et al., 27 Aug 2025).
CMP and surface treatments: For ScAlN, CMP reduces roughness from ~5 nm to <1 nm, dramatically lowering scattering loss (Wang et al., 2024).

3. Waveguide Geometry, Mode Confinement, and Dispersion Control

Device geometry is optimized for minimal modal overlap with rough regions, tight field confinement, low bending/radiation loss, and targeted dispersion.

Cross-section: Standard single-crystal AlN UV microrings employ H=400 nm, W=600 nm with sapphire cladding; for telecom, 1.1 μm-thick fully-etched ridges are common (Liu et al., 2018, Bruch et al., 2018).
Index contrast: n_AlN(390 nm) ≃ 2.15, n_sapphire ≃ 1.75, n_SiO₂ ≃ 1.46, allows single-mode TE guidance with high transverse confinement (Γ_core ≃ 85%) and negligible substrate leakage (Liu et al., 2018, Xiong et al., 2014).
Bending loss: With radius R≥50 μm, ring loss <0.001 dB/90° is achieved (negligible impact) (Liu et al., 2018).
Slab/ridge structures: Telecom devices use partially-etched (W=1.8 μm, etched depth 800 nm of 1 μm) quasi-TM ridges, sometimes with residual 200-nm slab for dispersion engineering (Yan et al., 27 Jun 2025).
Mode profile simulation: COMSOL/Lumerical FEM solvers yield n_eff ≈ 1.98 at 390 nm (UV) and ~1.75–1.98 at telecom (Liu et al., 2018, Yan et al., 27 Jun 2025).
Multilayer and hybrid integration: Monolithic integration with AlGaN quantum wells or hybrid overlay of Si₃N₄/ScAlN enables additional device functionality (ferroelectricity, strong χ²) at no loss penalty (Liu et al., 1 Aug 2025).

4. Loss Mechanisms and Quantitative Analysis

Propagation loss in AlN waveguides emerges from a combination of sidewall Rayleigh scattering, defect/bulk absorption, and, in some regimes, substrate or void scattering.

Rayleigh scattering: α_scatt ∝ σ²/λ³ (or ∝ λ⁻⁴ for volumetric mechanisms), where σ is rms sidewall roughness. Measured as 0.08 dB/cm (σ=2 nm, λ=390 nm); at longer λ, α scales down (Liu et al., 2018, Videnov et al., 27 Aug 2025, Brunetta et al., 7 Nov 2025).
Bulk absorption: α_abs ≲ 0.01 dB/cm (UV), <0.001 dB/cm (telecom). Point defects and threading dislocations (10⁸ cm⁻²) set the absorption limit (Liu et al., 2018).
Void-induced loss: Incomplete buffer engineering or poor MOVPE growth introduces voids (height 20–300 nm) at AlN/sapphire interface; density N=120 μm⁻² and diameter >10 nm raise α >30 dB/cm (telecom) or >8 dB/cm (visible). Complete elimination via sputtered buffer reduces α <0.2 dB/cm (Brunetta et al., 7 Nov 2025).
Anneal and passivation: High-temperature anneal (900°C, N₂) passivates dangling bonds; ALD forms a conformal chemical barrier (Liu et al., 2018, Videnov et al., 27 Aug 2025).
Loss measurement/analysis: Direct cut-back (dB/cm) and microring Q-factor methods (Q_i = 2π n_g / (λ α)) yield α down to 0.12 dB/cm (UV), 0.154 dB/cm (telecom C-band), and 2 dB/cm at 852 nm (Liu et al., 2018, Singh et al., 2024, Videnov et al., 27 Aug 2025).

5. Platforms, Device Demonstrations, and Performance Benchmarks

AlN’s low-loss waveguides underpin a range of device demonstrations, including microring and photonic crystal cavities, second- and third-harmonic generation, and on-chip nonlinear optics.

UV/visible devices: Intrinsic Q_i≈4.3×10⁵ at 390 nm (0.12 dB/cm loss), single-mode TE₀ operation, full integration with AlGaN UV emitters (Liu et al., 2018).
Telecom devices: Sputtered polycrystalline AlN yields 0.154 dB/cm at 1550 nm, c-axis epitaxial films reach 0.08–0.36 dB/cm, intrinsic Q up to 1.0×10⁶ (Singh et al., 2024, Yan et al., 27 Jun 2025).
Nonlinear optics: SHG efficiency of 17,000%/W in phase-matched fully-etched microresonators (1.1 μm-thick, fully-etched, Q_0>10⁶), with pump-depletion at 3.5 mW (Bruch et al., 2018).
Photonic crystal nanobeams: Sputtered AlN-on-insulator cavities (330 nm) support Q up to 146,000, with α ≈2.6 dB/cm (Pernice et al., 2012).
ScAlN/Si₃N₄ hybrid: Confines light in low-loss Si₃N₄ (α = 1.03 dB/cm, Q_i = 3.35×10⁵), with functional ScAlN underlayer (Liu et al., 1 Aug 2025).
Loss-optimized visible: ALD+RTA on AlN-on-sapphire decreases α from ~7.3 dB/cm (baseline) to 2.0±0.3 dB/cm at 852 nm (Videnov et al., 27 Aug 2025).
Device yield and scaling: Full MOCVD runs on 2″ wafers demonstrated <10% thickness and resonance variation (Liu et al., 2018).

6. Critical Process-Performance Relationships and Future Directions

Film growth: Void elimination and single-crystallinity (sputtered buffer plus high-T MOCVD or MOCVD alone with careful nucleation) are essential to minimize volumetric scattering and mid-gap defect absorption, especially for visible/UV operation (Brunetta et al., 7 Nov 2025, Liu et al., 2018).
Pattern transfer: Single hard mask (Si₃N₄), no metal lift-off, and resin-based charge control reduce roughness below 1 nm and avert metal-induced loss (Yan et al., 27 Jun 2025).
Etch and passivation: Optimized RIE (low damage, near-vertical sidewalls), with post-fabrication ALD and RTA, outperform solely geometric confinement approaches at sub-micron wavelengths (Videnov et al., 27 Aug 2025).
Scandium doping: ScAlN enables advanced functionalities but presently at increased loss (~2.4 dB/cm) unless hybridized with Si₃N₄ (Wang et al., 2024, Liu et al., 1 Aug 2025). Improvements require better crystallinity, further roughness reduction, or thinner core design.
Loss scaling and forecast: Rayleigh and void scattering scale as λ⁻⁴. UV and visible PICs demand sub-nm RMS roughness, void-free epilayers, and advanced surface passivation not typically required for telecom (Brunetta et al., 7 Nov 2025, Videnov et al., 27 Aug 2025).
Yields: Well-controlled processes deliver >75% device yield with ≤0.01 dB/cm variation (C/O-band) (Singh et al., 2024).

A plausible implication is that as device scaling pushes toward deeper UV, improvements in void control, sidewall smoothness, and interface passivation—especially through buffer layer engineering and surface treatments—will define the achievable floor for loss in AlN photonic integrated circuits. Further integration with active materials, tighter bends, and multilayer hybrid stacks will rely on the continued refinement of these core fabrication methodologies.