Inverse-designed photonic interfaces beyond eigenmode expansion limits

Technical Summary: Inverse-designed photonic interfaces beyond eigenmode expansion limits

Problem Statement
Photonic integrated circuits (PICs) offer significant advantages in performance and scalability but suffer from severe mode mismatch when interfacing with standard optical fibers or other chips. Conventional solutions rely on waveguide tapers based on eigenmode-expansion (EME) theory, which adiabatically expand optical modes. However, this approach imposes rigid constraints: it requires long taper lengths (hundreds of microns) to ensure low-loss evolution and necessitates extremely small waveguide tips (often <300 nm) to match fiber mode field diameters (MFDs). These requirements frequently demand multi-layer tapering structures and multiple high-precision lithography steps, particularly on emerging platforms like thin-film lithium niobate (TFLN). In TFLN, the typical half-etched rib structure and non-vertical sidewalls (45–70°) make achieving large MFDs in a single layer difficult, often pushing the optical mode into the slab rather than expanding it effectively. Consequently, existing coupling strategies are often bulky, fabrication-intensive, and limited in bandwidth.

Methodology
The authors propose a topology optimization approach using inverse design to overcome the limitations of traditional EME-based tapers. Instead of sweeping cross-sectional geometries to match eigenmodes, the interface is treated as a "black box" where the coupling efficiency is optimized directly through a transient field evolution process.

• Design Strategy: A Gaussian beam with a predetermined MFD is injected into a design region. The algorithm optimizes the distribution of dielectric material (rib vs. slab) within this region to maximize the fundamental TE₀ mode power at the output waveguide port.
• Simulation & Constraints: The optimization utilizes 3D finite-difference time-domain (FDTD) simulations via the MEEP software package with gradient-based topology optimization. Crucially, the design incorporates practical fabrication constraints specific to TFLN, including etching depth (300 nm), sidewall angles (60°), crystal anisotropy, and a minimum feature size of 400 nm (compatible with i-line stepper lithography).
• Optimization Process: The process involves three phases: continuous optimization, geometric constraint enforcement for minimum feature size, and sidewall feature projection. The design region is initialized with a uniform material density to ensure an unbiased starting point.

Key Contributions

1. Transcending EME Limits: The paper demonstrates that inverse-designed structures can achieve spot-size conversion through transient field evolution and constructive interference, rather than relying solely on adiabatic eigenmode expansion. This allows for effective mode expansion in both vertical and horizontal dimensions within a single-layer configuration.
2. Ultra-Compact, Single-Layer TFLN Interfaces: The authors successfully designed interfaces for TFLN that are fully compatible with standard rib waveguide components. Unlike previous methods requiring multi-layer overlays, these designs require only a single high-resolution lithography step.
3. Broadband Performance: The designs achieve low coupling loss with exceptionally broad operational bandwidths, exceeding 120 nm within the 1-dB loss window.
4. Generalizability: The methodology is shown to be adaptable to various scenarios, including coupling to standard single-mode fibers (SMF), III-V chips (normal and oblique incidence), and visible wavelengths, demonstrating platform-agnostic potential.

Experimental Results
The authors fabricated and characterized the inverse-designed interfaces on x-cut TFLN wafers.

• Fabrication: Devices were patterned using an ASML UV stepper, followed by reactive ion etching (RIE) to create 300 nm deep ribs. A second lithography step was used to define the chip facets, and XeF₂ isotropic etching was employed to remove the silicon substrate beneath the couplers to create suspended structures.
• Coupling Efficiency:
  • Lensed Fiber (MFD = 2 µm): Measured coupling loss of 3 dB/facet (simulated ~1 dB), representing a 4.4 dB improvement over a bare waveguide.
  • Ultra-High Numerical Aperture (UHNA) Fiber (MFD = 3.2 µm): Measured coupling loss of 3.5 dB/facet (simulated ~1 dB).
• Bandwidth: Both designs exhibited a 1-dB bandwidth exceeding 120 nm (1520–1640 nm).
• Robustness: Robustness analysis indicated that the designs remain effective despite fabrication variations of ±43 nm (erosion/dilation), with simulated penalties remaining below 0.25 dB.

Significance and Claims
The paper claims to present a "brand-new photonic interfacing strategy" that offers superior flexibility for interfacing various fiber types and chip-to-chip connections. By moving beyond traditional eigenmode evolution theories, the proposed inverse-designed interfaces significantly reduce device footprints (by an order of magnitude) and streamline fabrication by eliminating the need for complex multi-layer processes. The authors assert that this approach is broadly applicable to a variety of material platforms (including silicon, diamond, and silicon carbide) and application scenarios, such as optical interconnects, telecommunications, and on-chip nonlinear optics. The work addresses a critical bottleneck in TFLN PICs—the lack of effective, low-cost, and fabrication-friendly interfaces—thereby enabling the full potential of high-speed modulation, frequency comb generation, and nonlinear frequency conversion on this platform.