Inverse-Designed Grating Couplers with Tunable Wavelength via Scaling and Biasing

TL;DR

This work tackles the vulnerability of inverse-designed grating couplers to fabrication-induced spectral shifts by introducing a post-design correction strategy based on systematic scaling and edge biasing. Using adjoint topology optimization with broadband objectives, the authors design compact, broadband GC devices on a silicon-on-insulator platform and validate them with wafer-scale fabrication and 4f-scanning measurements. The key finding is that applying scale factors and contour biases can recover substantial efficiency and tune the operating wavelength, achieving up to eightfold improvement relative to unscaled/unbiased cases and enabling practical wafer-scale testing and cryogenic operation. The approach delivers a robust, transferable framework for inverse-designed nanophotonic devices, with potential extensions to mode, wavelength, or power splitters and broad applicability in next-generation photonic and quantum technologies.

Abstract

Photonic integrated circuits are heavily researched devices for telecommunication, biosensing, and quantum technologies. Wafer-scale fabrication and testing are crucial for reducing costs and enabling large-scale deployment. Grating couplers allow non-invasive measurements before packaging, but classical designs rely on long tapers and narrow bandwidths. In this work, we present compact, inverse-designed grating couplers with broadband transmission. We optimized and fabricated arrays of devices and characterized them with a 4f-scanning setup. The nominal design reached simulated efficiencies of 52 %, while measurements confirmed robust performance with up to 32 % efficiency at the target 1540 nm wavelength and 46 % at shifted wavelengths. Without scaling and contour biasing, the measured efficiency at the target wavelength drops to only 4.4 %. Thus, a key finding is that systematic scaling and edge biasing recover up to an eightfold improvement in efficiency. These inverse-designed grating couplers can be efficiently corrected post-design, enabling reliable performance despite fabrication deviations. This approach allows simple layout adjustments to compensate for process-induced variations, supporting wafer-scale testing, cryogenic photonic applications, and rapid design wavelength tuning.

Figures (4)

• Figure 1: Illustration of scaling and biasing applied to a simple donut structure. Scaling preserves the design proportions, while biasing shifts the contour inward or outward to emulate fabrication over- and underetching.
• Figure 2: (a) Photonic test device consisting of two GC connected by a 100µm-long waveguide for coupling efficiency characterization. (b)*SEM image of the nominal design with GDSII mask overlay. Small detached islands are visible but have negligible impact since the optical field vanishes at the coupler edges. (c) Cleaved sidewall showing footing at the Si–SiO2 interface caused by the RIE step.
• Figure 3: Simulated and measured efficiency spectra of GC. (a) Simulation for fixed scale (100%) and varying edge bias. (b) Simulation for fixed edge bias (0nm) and varying scale. (c) and (d) Measured spectra from the 4f-scanning setup for the same parameters, shown as dotted raw data and solid smoothed curves. Thick gray spectra indicate the nominal design with 0nm edge bias or 100% scale. The gray area indicate where the measured data was truncated.
• Figure 4: Simulation (a–c) and measurement (d–f) results of scaled and biased GC. White cells mark the experimentally tested devices, while the rest are obtained by linear inter- or extrapolation. (a) and (d) Center wavelength extracted from the FWHM of the efficiency spectrum. (b) and (e)FWHM bandwidth of the efficiency spectra. (c) and (f) Efficiency at the design wavelength of 1540nm.