Inverse-Designed Silicon Nitride Nanophotonics

Technical Summary: Inverse-Designed Silicon Nitride Nanophotonics

Problem Statement
Silicon nitride (Si$_3$N$_4$) photonics has become a mature platform for integrated circuits due to its wide transparency window, low linear and nonlinear losses, and high mode confinement. However, traditional photonic design approaches often rely on intuitive, parameterized geometries that limit the exploration of complex device functionalities. While inverse design—using gradient-based optimization to achieve desired performance—has been successfully demonstrated on silicon, diamond, silicon carbide, and lithium niobate platforms, experimental validation of inverse-designed nanophotonics on thick, high-quality silicon nitride platforms has remained elusive. Furthermore, the lower refractive index contrast of Si$_3$N$_4$ compared to silicon necessitates larger design areas, requiring benchmarks on how optimization performance scales with footprint size.

Methodology
The authors employed a gradient-based inverse design algorithm (utilizing the SPINS software framework) to optimize photonic structures on a foundry-compatible, thick silicon nitride platform. Two distinct fabrication processes were used:

1. Reactive Magnetron Sputtering: Used for devices with varying thicknesses (400 nm, 730 nm, and 790 nm).
2. Low-Pressure Chemical Vapor Deposition (LPCVD): Used for devices with a uniform 800 nm thickness.

The optimization process incorporated fabrication constraints, including a minimum feature size of 80–100 nm and moderate curvature tolerance, to ensure manufacturability. A 40 nm square mesh was used for accurate simulation. Post-optimization, designs were verified using full 3D Finite-Difference Time-Domain (FDTD) simulations (Lumerical), and small features prone to fabrication errors were manually removed.

Three primary device classes were designed, fabricated, and characterized:

• Wavelength-Division Multiplexers (WDMs): 3-port devices separating broadband inputs into spectrally distinct channels.
• Mode-Division Multiplexers (MDMs): Devices separating orthogonal spatial modes (TE$_{00}$ and TE$_{10}$) into separate waveguides.
• Inverse-Designed Reflectors: Used to construct Fabry-Pérot (FP) microresonators with high reflectivity and mode filtering capabilities.

Key Contributions and Results

• Wavelength-Division Multiplexers (WDMs):

  • Two devices were demonstrated: one operating at 194/232 THz with a $5\times5$ $\mu$m$^2$ footprint, and another at 196/202 THz with an $8\times8$ $\mu$m$^2$ footprint.
  • Performance: Both devices exhibited insertion losses of approximately -2 dB at their central wavelengths. Crosstalk levels were measured below -10 dB (specifically -7 dB in the C-band and -3 dB in the O-band for the 194/232 THz device; -11 dB for the 196/202 THz device).
  • Bandwidth: Simulated 3-dB bandwidths exceeded 30 THz for both frequency bands.
  • Benchmarking: The study compared performance across different optimization areas, demonstrating that larger footprints (e.g., 8x8 $\mu$m$^2$) allowed for closer channel spacing (196/202 THz) while maintaining performance.

• Mode-Division Multiplexers (MDMs):

  • Devices were designed to route TE$_{00}$ and TE$_{10}$ modes from a multi-mode waveguide into single-mode outputs.
  • Footprint Comparison: An $8\times8$ $\mu$m$^2$ device achieved a crosstalk of -18 dB at 194.5 THz. A larger $15\times10$ $\mu$m$^2$ device (using 800 nm thick film) reduced insertion loss to sub -1 dB and significantly improved crosstalk stability across the C- and L-bands (variation of only -1 dB compared to -15 dB for the smaller device).
  • Insertion Loss: Single MDM devices showed average insertion losses around -2 dB.

• Inverse-Designed Reflectors and Microresonators:

  • Reflectors were optimized for multi-mode cavity waveguides to reduce interaction with scattering sidewalls. A design area of $11\times2.8$ $\mu$m$^2$ was used.
  • Reflectivity: The reflectors achieved average reflectivity above 97% across a 3 THz bandwidth, with a maximum of 98.5% for the TE$_{00}$ mode family.
  • Resonator Performance: Fabry-Pérot cavities formed by sandwiching a waveguide between two such reflectors demonstrated a maximum loaded finesse of 162 (corresponding to a Q-factor of ~210,000) at ~190.7 THz.
  • Platform Characterization: Ring resonators fabricated on the same chip (using 800 nm thick Si$_3$N$_4$) exhibited intrinsic Q-factors over 20 million and propagation losses of approximately 8 dB/m.

Significance and Claims
The paper claims to provide the first experimental demonstration of inverse-designed photonics on a thick, high-quality silicon nitride platform. The work demonstrates that inverse design can successfully expand the photonic design library beyond traditional approaches, enabling precisely tailored wavelength-division multiplexers, mode-division multiplexers, and high-Q resonators.

The authors emphasize that the thick (800 nm) silicon nitride platform is critical for achieving compact nonlinear and quantum photonic operations, as it offers a balance of high mode confinement and low loss that thinner films cannot provide. The results suggest that inverse-designed structures on this platform are robust against fabrication errors and are readily integrable into nonlinear and quantum photonic circuits. The study concludes that these advancements pave the way for scalable, high-performance integration in silicon nitride-based systems, particularly for applications requiring dispersion engineering and compact nonlinear interactions.