Sputtered AlN buffer layer for low-loss crystalline AlN-on-sapphire integrated photonics

TL;DR

This work identifies voids in AlN-on-sapphire epilayers as a dominant source of scattering losses at telecom wavelengths, showing that their size and density strongly govern propagation losses via FDTD simulations. A thin sputtered AlN buffer enables void-free hybrid epilayers that achieve intrinsic quality factors approaching $2.0×10^6$ and propagation losses below $0.2~dB~cm^{-1}$ at $1550~nm$, enabling high-performance linear and nonlinear photonics. The authors demonstrate SHG and broadband SCG in dispersion-engineered WGs on these void-free layers, and show the substantial improvement over void-containing layers. The results suggest a viable path to low-loss visible/UV AlN photonics via the hybrid approach, with broad implications for nonlinear PICs and integrated quantum photonics.

Abstract

In recent years, aluminum nitride (AlN) has emerged as an attractive material for integrated photonics due to its low propagation losses, wide transparency window, and presence of both second- and third-order optical nonlinearities. However, most of the research led on this platform has primarily focused on applications, rather than material optimization, although the latter is equally important to ensure its technological maturity. In this work, we show that voids, which are commonly found in crystalline AlN-on-sapphire epilayers, have a detrimental role in related photonic structures, as they can lead to propagation losses exceeding 30 dB cm$^{-1}$ at 1550 nm. Their impact on light propagation is further quantified through finite-difference time-domain simulations that reveal that void-related scattering losses are strongly dependent on their size and density in the layer. As a possible solution, we demonstrate that when introducing a thin sputtered AlN buffer layer prior to initiating AlN epitaxial growth, void-free layers are obtained. They exhibit intrinsic quality factors in microring resonators as high as $2.0\times 10^6$, corresponding to propagation losses lower than 0.2 dB cm$^{-1}$ at 1550 nm. These void-free layers are further benchmarked for high-power applications through second-harmonic and supercontinuum generation in dispersion-engineered waveguides. Such layers are highly promising candidates for short-wavelength photonic integrated circuit applications, particularly given the strong potential of AlN for visible photonics. Given that volumetric scattering losses scale as $λ^{-4}$, the platform quality becomes increasingly critical in the visible and ultraviolet range, where our improved layers are expected to deliver enhanced performance.

Figures (8)

• Figure 1: (a)-(c) Cross-section SEM images of the epilayers A, B and C, respectively, with a high magnification inset for the AlN-sapphire interface region. Samples are coated with a thin metal layer for charge-dissipation purposes. (d)-(f) Surface AFM scans of the epilayers A, B and C, respectively, after thinning them down by ICP dry etching to a thickness of 110nm (horizontal dashed orange lines in the SEM pictures).
• Figure 2: (a) to (c) SIMS concentration profiles of the main impurity species for the three AlN-on-sapphire epilayers under investigation. The higher hydrogen concentration level measured in sample A is consistent with the presence of voids in the first 400nm of the layer.
• Figure 3: (a) False-color top view SEM image of an AlN MRR of 60µm radius prior to SiO2 cladding deposition. (b) Enlarged bird's eye view of an MRR showing a smooth sidewall. (c) False-color cross-section SEM image of a bus WG facet in sample C, revealing a sidewall angle of about 80.
• Figure 4: (a) Optical transmission in 1.4µm-wide WGs fabricated from sample A, characterized at 1550 nm in TM polarization. (b) Transmission spectrum of a TM00 resonance in a 2.3µm-wide MRR from sample A, showing optical losses around 30dB. (c) to (f) Characterization of the fundamental TE and TM resonances in the C-band for 1.8µm-wide MRRs fabricated from samples B and C. Rings have a radius of 60µm and measurements were done in the undercoupled regime.
• Figure 5: Experimentally measured mean $Q_\text{int}$ value of the fundamental TE and TM resonances at $\lambda\simeq1550nm$ in MRRs fabricated from sample B. The error bars correspond to the standard deviation for each MRR set. Microrings have a radius of 60µm, an upper width that is varied from 1.2to 2.9µm, and a bus WG to ring resonator coupling gap chosen to ensure that MRRs operate in the undercoupled regime.