J. Eur. Opt. Society-Rapid Publ. 22, 56( 2026) 547
Fig. 4. Per-bend loss for the TE a) and TM b) polarization, comparing standard bends to laterally shifted bend geometries across different bend radii. In the shifted designs, K denotes the lateral offset, as illustrated in the inset of a). c) Measured attenuation of cascaded MMI splitters for the TE and TM modes. The linear fits provide the attenuation per MMI device and include additional contributions from bend and propagation losses.
suppressed in this regime. For TM polarization( Fig. 4b), a similar trend is observed. For bend radii of 20 lm and above, losses remain low, with values of approximately 0.04 dB for a 40 lm bend and as low as 0.02 dB for the 20 lm bend-shift design. As for TE polarization, the bend-shift geometry consistently outperforms the standard bend, confirming its effectiveness in reducing mode mismatch loss for both polarizations. At a bend radius of 10 lm, a pronounced increase in loss is observed for both polarizations, exceeding 0.10 dB per bend. In this regime, TE losses increase to approximately 0.11 dB, while TM losses reach about 0.16 dB, indicating the onset of radiation and mode mismatch losses due to insufficient mode confinement at tight bend radii.
2.3.2.4
MMI splitter
The splitter losses were evaluated using an MMI splitter tree terminated by grating couplers, each configuration containing a different number of MMIs. The total transmitted optical power was obtained from the summed pixel intensity px sum per exposure time s exp monitored by the CMOS camera and plotted logarithmically as a function of the number of MMIs. The extracted slope represents the combined loss per MMI arm of one MMI, two bends, and a 150 lm straight waveguide section. From this, the individual MMI insertion loss was determined to be( 1.88 ± 0.16) dB for TE and( 0.62 ± 0.30) dB for TM( see Fig. 4c).
2.3.2.5 Incoupling
The incoupling loss was estimated in two ways:( i) from propagation-loss measurements by subtracting the contributions from bends and EBL – OL taper sections from the y-intercept, and( ii) from direct transmission measurements through a straight waveguide across the chip, with propagation losses subtracted. Method( i) yields single-sided incoupling losses of( 13.0 ± 1.5) dB for TE and( 14.08 ± 0.81) dB for TM, while method( ii) gives( 13.89 ± 0.43) dB for TE and( 14.71 ± 0.43) dB for TM( see Fig. 3b). These incoupling loss values correspond to the sum of the fiber-to-chip edge-coupling loss and the loss introduced by the two-layer adiabatic coupler. The total incoupling loss per facet is given by these contributions, including additional EBL-OL taper loss. The measured losses are significantly higher than those predicted by simulation, which may be attributed to fabrication-induced deviations in the taper-tip widths.
3 Design and characterization of the grating couplers
The operational principle of the structure is illustrated schematically in Figure 5a. Two beams corresponding to orthogonal TE and TM polarizations are emitted from a single GC and through an opening in the gold electrode layer. An optical microscope image of the fabricated device is presented in Figure 5b.
3.1 Design
The design optimization was performed using a combination of 2D and 3D finite-difference time-domain( FDTD) simulations via Ansys Lumerical. In the 2D model, the grating periods and duty cycles were varied to control the emission angle and far-field beam shape for the TE polarization [ 49 – 51 ]. To finalize the design, a 3D FDTD simulation was used to define the grating taper geometry and grating taper width and to extract the far-field intensity profile from the resulting structure. The GC design is primarily optimized for TE operation, resulting in a 30 lm-long grating section for efficient TE outcoupling, while simultaneously supportingawell-defined TM-polarized response, enabling a comprehensive assessment of polarization selectivity, angular separation of the emitted beams, efficiency and their quality.
3.2 Measurement
The outcoupled intensity distribution was recorded slice by slice in the xy-plane by translating the microscope system along the z-direction with a step size of 0.5 lm. At each z- slice, the intensity profile of the emitted beam was