J. Eur. Opt. Society-Rapid Publ. 22, 56( 2026) 549
Fig. 6. Simulated( a, c) and measured( b, d) far-field intensity distributions of the laser-beam waist at 100 lm above the chip surface for TE( a, b) and TM polarization( c, d).
oxide layers to suppress. Transitioning to fused silica substrates would further mitigate substrate-related losses [ 52 ].
4.2 Incoupling
Incoupling losses of around 14 dB exceed the simulated values, yet both evaluation methods yielded similar absolute coupling efficiencies, with TE coupling slightly lower than TM. For comparison, visible-wavelength dual-layer Si 3 N 4 couplers have been reported with losses of about 4 dB / facet [ 38 ], while single-layer couplers typically reach around 7.5 – 8.5 dB / facet [ 20 ]. The values obtained here are higher than those reported, suggesting that additional loss mechanisms are present in the fabricated devices. This behavior is plausibly linked to the two-layer edge-coupler architecture, which is highly sensitive to the taper-tip width of the thick-core Si 3 N 4 layer [ 38 ]. A taper-tip wider than the optimized target value reduces the adiabaticity of the local index transition near the taper entrance and diminishes modal overlap( see Supplementary material). This indicates that future designs require an improved fabrication of the taper tip, or a thinner thick-core Si 3 N 4 layer to reduce sensitivity to the taper-tip width. Measurements with the fiber array yielded coupling efficiencies comparable to the singlefiber configuration, indicating that alignment constraints introduce no significant additional penalty.
4.3 Bend and bend-shift losses
The measured bend losses of around 0.04 dB per 20-lmradius bend match results reported in visible-wavelength Si 3 N 4 experiments [ 20 ]. TM-polarized light exhibited noticeably higher loss for small-radius bends, a consequence of the 200 nm Si 3 N 4 thickness, which was optimized for TE operation and provides insufficient confinement for TM, thereby increasing curvature-induced radiation loss. Such polarization-dependent bend performance aligns with earlier analyses of thin-core Si 3 N 4 waveguides [ 19 ]. The shift of the bendloss minima to larger radii – from 20 lm for TE to 40 lmfor TM modes – is consistent with the lower propagation loss observed for TM.
Introducing bend-shift structures proved particularly beneficial for tight bends, where they reduced excess loss associated with mode mismatch at straight-to-bend transitions. This strategy is valuable for compact photonic-circuit layouts or applications requiring minimal stray light and crosstalk, as it enables dense routing without compromising optical efficiency.
4.4 MMI splitter
Significant deviations from simulated values were observed for TE MMI splitter insertion losses, while TM performance remained close to predictions. Enhanced TE losses can be attributed to stronger sensitivity to sidewall scattering within the multimode region, which perturbs the self-imaging condition. This observation is consistent with the higher TE propagation losses across the platform. TM modes, with weaker confinement, are less susceptible to such imperfections. Further optimization of the MMI length would likely reduce the residual insertion loss( see Supplementary material).
4.5 Taper losses
The OL-EBL taper transitions show good agreement with simulations for the TE polarization, while larger deviations are observed for the TM mode. These deviations may arise from lateral and longitudinal misalignment, as well as propagation losses, which are not accounted for in the simulations. Additionally, the observed behavior can be attributed to polarization-dependent changes in the modal field distribution induced by the taper. For the TE mode, tapering reduces the overlap with the etched sidewalls, thereby lowering sidewall-scattering loss. In contrast, for the TM mode, tapering may enhance propagation loss by increasing the overlap of the modal field with horizontal layer interfaces, thereby amplifying interface-related loss mechanisms. Extending the taper length or employing higher-order polynomial or spline profiles could further improve adiabaticity and reduce insertion loss.
4.6 Grating coupler
The measured grating emission profiles closely match the simulated far-field profiles in both overall shape and divergence. When quantified using Gaussian beam waists along the fitted principal axes, the average absolute deviation between simulated and measured data is 17 – 18 % for both polarization regimes. The TM profiles exhibit a side lobe originating from the substrate back-reflected beam, which becomes visible because the lateral offset between the upward- and downward-reflected beams increases at larger angles. The measured emission angles are likewise consistent with the simulated values: for TE polarization, the emission angle matches the simulated value and agrees within experimental uncertainty, whereas for TM polarization, a deviation of approximately 4 ° is observed. These