Three-Wave Interaction Grating Coupler with Sub-Decibel Insertion Loss at Normal Incidence

TL;DR

The paper tackles the challenge of achieving high-efficiency, normal-incidence coupling into silicon photonics without relying on edge coupling or extra material layers. It introduces three-wave interaction gratings (TWIGs) and optimizes their 2D dielectric geometry via adjoint inverse design (EMOPT) to maximize peak efficiency and bandwidth while meeting fabrication constraints, predicting a $91\%$ peak efficiency and a $34$ nm $1$ dB bandwidth. Fabricated in a commercial foundry, the focusing TWIG delivers $85.4\%$ coupling at $1546.4$ nm with a $20$ nm bandwidth, and the 1D TWIG delivers $81.1\%$ at $1555.5$ nm with the same bandwidth, corroborating the simulations and marking the highest normal-incidence grating-coupler performance on a $220$ nm SOI platform without additional layers. The work enables efficient, vertically oriented coupling to components like VCSELs and multicore fibers and provides a scalable, wafer-friendly path toward broader adoption of grating couplers in silicon photonics.

Abstract

We report the design, fabrication in a commercial foundry, and experimental results of high-efficiency, normal incidence grating couplers for silicon photonics. We observe a maximum coupling efficiency of 85.4% (-0.69 dB) with a 1 dB bandwidth of 20 nm at a central wavelength of 1546 nm. These experimental results verify earlier theoretical and simulation results and pave the way for the use of perfectly vertical grating couplers, as an alternative to edge coupling, in silicon photonics applications that are sensitive to input coupling loss. Further, these results enable the use of grating couplers for vertically oriented elements, such as multicore fibers and VCSELs, and address challenges associated with coupling to free space beams.

Figures (8)

• Figure 1: Generalized geometry of a three-wave interaction grating parameterized by $W_{n,m}$, $\Lambda_{n,m}$, $H_m$, $d_1$ and $d_2$. $W_{n,m}$ represents the width of the $m^{th}$ scatterer in the $n^{th}$ local period. $\Lambda_{n,m}$ represents the separation between the $m^{th}$ scatterers in the $n$ and $n+1$ local periods. $H_m$ is the etch depth of the $m^{th}$ scattering site across all local periods. $d_1$ and $d_2$ are initial offset values between the scattering sites in the first local period. Panels a and b show the back reflected and downward scattered waves from three scattering sites respectively.
• Figure 2: Simulation results of TWIG coupler optimization. (a) Optimized geometry demonstrating apodization of the etch widths to produce a gaussian mode profile. The background index here is associated with silicon dioxide (1.444). (b) Electric field profile associated with the quasi-TE mode of the grating. The output is directed perfectly vertical with minimal back reflections and substrate losses. (c) Coupling spectra of the optimized coupler with a peak efficiency of 91 $\%$ and a 1 dB bandwidth of 34 nm spanning the c-band. Field profiles generated by a (d) one-dimensional and (e) focusing TWIG coupler.
• Figure 3: Simulated impact of fabrication variations for (a) etch width bias, (b) etch depth error of the partially etched scattering sites, and (c) lateral alignment error between the partially and fully etched scattering sites. Etch width bias results in a reduction to the peak efficiency, a shift in the center wavelength, and for positive bias a broadening of the spectrum of a TWIG coupler. The dominant effect of an etch depth error is a shift in the central wavelength. These optimized devices are robust to alignment error and experience little affect over a wide range of errors.
• Figure 4: SEM images of the focusing TWIG coupler provided by Applied Nanotools.
• Figure 5: (a) Transmission and (b) reflection spectra of the highest performing one-dimensional TWIG coupler. A waveguide propagation loss of -1.35 dB/cm is estimated from an analysis of the Fabry-Perot resonances.