SPIE Journal Paper | 27 April 2022
KEYWORDS: Waveguides, Silicon, Lithium, Silica, Finite-difference time-domain method, Electron beam lithography, Integrated optics, Optimization (mathematics), Lithography, Binary data
Compact, multimode, silicon waveguide bends are essential building blocks for high-density, integrated-optics, mode-division-multiplexing (MDM) circuits. However, sharp, multimode, silicon waveguide bends can typically introduce considerable insertion losses and inter-mode crosstalk. Significant efforts have been made to realize compact and high performance multimode silicon waveguide bends, but most of those designs either require complicated fabrication processes or are not really compact. We show that the transmission performances of a regular, ultra-sharp, multimode, silicon waveguide bend can be drastically improved by directly optimizing its topology. The optimization can be efficiently accomplished by using an adjoint-based inverse design. Through this design strategy, we realize a three-mode, silicon waveguide bend that is ultra-compact, with a small radius of 5.5 μm, while having ultra-low average insertion losses of 0.023, 0.046, and 0.068 dB and small average mode crosstalk of −26, −27, and −27 dB for the input modes of TE0 − 2 (TE = transverse electric), respectively, in the wavelength range 1500 to 1600 nm. Also, the waveguide bend can be fabricated through a single electron-beam lithography step. The significant advantages in size and performance of the current waveguide bend show the great potential of the proposed inverse design strategy for designing multimode waveguide bends for MDM systems.
