Photonic circuits find applications in biomedicine, manufacturing, quantum computing and communications. Photonic waveguides are crucial components, typically having cross-section orders of magnitude inferior when compared with other photonic components (e.g., optical fibers, light sources and photodetectors). Several light-coupling methods exist, consisting of either on-plane (e.g., adiabatic and end-fire coupling) or off-plane methods (e.g., grating and vertical couplers). The grating coupler is a versatile light-transference technique which can be tested at wafer level, not requiring specific fiber terminations or additional optical components, like lenses, polarizers or prisms. This study focuses on fully-etched grating couplers without a bottom reflector, made from hydrogenated amorphous silicon (a-Si:H), deposited over a silica substrate. Different coupler designs were tested, and of these we highlight two: the superimposition of two lithographic masks with different periods and an offset between them to create a random distribution and a technique based on the quadratic refractive-index variation along the device’s length. Results were obtained by 2D-FDTD simulation. The designed grating couplers achieve coupling efficiencies for the TE-like mode over −8 dB (mask overlap) and −3 dB (quadratic variation), at a wavelength of 1550 nm. The coupling scheme considers a 220 nm a-Si:H waveguide and an SMF-28 optical fiber.

Photonic circuits are an enabling technology for the development of novel solutions in different fields such as healthcare, quantum computing, neural networks, communications, and manufacturing. Interconnections between devices and systems require low-loss light coupling strategies. Grating couplers are a promising solution to couple light between photonic circuits and optical fibres due to their off-plane coupling capabilities. Hydrogenated amorphous silicon (a-Si:H), which can be deposited by PECVD over a substrate of silica or glass, is a suitable low-cost solution for the production of such light coupling devices. In this work we developed, trained and tested a fully connected feedforward neural network for coupling efficiency prediction in a-Si:H grating couplers. The light coupling gratings were simulated by twodimensional finite-difference time-domain (FDTD) analysis and field distributions were analysed with the Finite Element Method (FEM). Simulated gratings include non-apodized, linear and quadratic refractive index variation designs featuring full or partial etching, operating at 1550 nm. Not featuring any type of bottom reflector, the couplers exhibit coupling efficiencies up to about 40 % (~ -4 dB). The neural network multiclass grating coupler efficiency classifier was trained with over 3000 simulation results, reaching an accuracy over 85%, for coupling efficiencies between 0 and 30%+.