J. Eur. Opt. Society-Rapid Publ. 22, 56( 2026) 545
The experimental characterization reported here is performed on representative photonic test structures fabricated only up to layer L-5, before the deposition of the metallic layers.
2.2 Photonic component design 2.2.1 Alignment-tolerant incoupling
The incoupling region consists of three functional parts employing two vertically separated Si 3 N 4 layers. The thinner upper layer serves for mode matching to the input fiber( see Fig. 2a), providing an expanded optical mode comparable to the fiber mode field diameter, while the thick-core lower layer serves as the routing layer for on-chip propagation. Light is transferred adiabatically between the two layers by a vertically tapered section( see Fig. 2b), which minimizes mode conversion loss through gradual effective index evolution [ 37 ]. Such dual-layer and vertically adiabatic coupling concepts are widely adopted in Si 3 N 4 and Si-based photonic platforms to achieve efficient fiber-tochip coupling, sub-dB loss, and relaxed alignment tolerances [ 38 – 46 ]. Finally, a second taper region is introduced to compensate for potential lateral misalignment between the EBL and OL process levels, ensuring robust and efficient coupling( see Fig. 2c).
The designs were carried out using Ansys Lumerical, employing the Finite Difference Eigenmode( FDE) solver for mode-overlap simulations to optimize fiber-to-chip coupling and eigenmode expansion( EME) simulations were used to analyze adiabatic transitions and lithography overlap sections. The simulated total incoupling loss is 1.95 dB for TE and 1.65 dB for TM, assuming perfect alignment across all fabrication steps. When considering a 500 nm lateral misalignment at each coupling stage, the total incoupling loss increases to 2.26 dB( TE) and 2.03 dB( TM), demonstrating the alignment-tolerance of the incoupling design. Resulting design parameters, coupling efficiencies, and simulations are summarized in the Supplementary material.
2.2.2 Routing
To deliver light to the ions, it is routed across the chip via WGs, bends, and splitters before being directed into the GC.
2.2.2.1 Waveguides
WG modes were simulated using the FDE solver and the final waveguide geometry was then fixed to 200 nm 520 nm( height width) to secure single-mode operation and avoid higher-order contributions in the GC emission( see Supplementary material).
2.2.2.2 Bends
Bend structures with radii of 10, 20, 40, and 60 lm were designed, along with bend-shift structures optimized for TE mode matching inbetween bent and straight WGs using the FDE solver [ 47 ]. Bend shifts K of 30, 10, 10 nm were implemented for the 10, 20, 40 lm bends, respectively. They offer a promising approach to lowering mode-mismatch
Fig. 2. Schematics of vertical fiber-to-chip coupling scheme.( a) Illustration of the fiber-to-chip edge coupling into the PIC.( b) Geometry of the two-layer adiabatic taper for efficient mode conversion between L-4 and L-2 waveguides.( c) Schematic of the EBL-to-OL coupling stage. The red arrows depict the light propagation through the structures.
losses in compact bends, which in turn increases the flexibility of the photonic design [ 47 ]. 2.2.2.3 Splitter multimode interference( MMI) splitters are common onchip components used for optical power splitting based on self-imaging effects in multimode waveguides [ 48 ]. The MMI splitters were designed using EME simulations for TE-polarized light( see Supplementary material). Each splitter features a core length of 38 lm, a core width of 6 lm, taper sections of 10 lm lengthwith1.6lm width, and an output channel separation of 3.14 lm.
2.3 Experimental comparison of TE and TM losses 2.3.1 Setup
A fiber-coupled 760 nm distributed Bragg reflector( DBR) laser 1 was used as the light source for the characterization of the PIC. The input polarization was adjusted, and 1 Thorlabs DBR760PN.