Nanophotonics finds ever broadening applications requiring complex components with many parameters to be simultaneously designed. Recent methodologies employing optimization algorithms commonly focus on a single performance objective, provide isolated designs, and do not describe how the design parameters influence the device behaviour. Here we propose and demonstrate a machine-learning-based approach to map and characterize the multi-parameter design space of nanophotonic components. Pattern recognition is used to reveal the relationship between an initial sparse set of optimized designs through a significant reduction in the number of characterizing parameters. This defines a design sub-space of lower dimensionality that can be mapped faster by orders of magnitude than the original design space. The behavior for multiple performance criteria is visualized, revealing the interplay of the design parameters, highlighting performance and structural limitations, and inspiring new design ideas. This global perspective on high-dimensional design problems represents a major shift in modern nanophotonic design and provides a powerful tool to explore complexity in next-generation devices.

These authors contributed equally to this work. AbstractNanophotonics finds ever broadening applications requiring complex component designs with a large number of parameters to be simultaneously optimized. Recent methodologies employing optimization algorithms commonly focus on a single design objective, provide isolated designs, and do not describe how the design parameters influence the device behaviour. Here we propose and demonstrate a machinelearning-based approach to map and characterize the multi-parameter design space of nanophotonic components. Pattern recognition is used to reveal the relationship between an initial sparse set of optimized designs through a significant reduction in the number of characterizing parameters. This defines a design sub-space of lower dimensionality that can be mapped faster by orders of magnitude than the original design space. As a result, multiple performance criteria are clearly visualized, revealing the interplay of the design parameters, highlighting performance and structural limitations, and inspiring new design ideas. This global perspective on high-dimensional design problems represents a major shift in how modern nanophotonic design is approached and provides a powerful tool to explore complexity in next-generation devices.

We present perfectly vertical grating couplers for the 220 nm silicon-on-insulator platform incorporating subwavelength metamaterials to increase the minimum feature sizes and achieve broadband low back-reflection. Our study reveals that devices with high coupling efficiencies are distributed over a wide region of the design space with varied back-reflections, while still maintaining minimum feature sizes larger than 100 nm and even 130 nm. Using 3D-finite-difference time-domain simulations, we demonstrate devices with broadband low back-reflection of less than − 20 d B over more than 100 nm bandwidth centered around the C-band. Coupling efficiencies of 72% and 67% are achieved for minimum feature sizes of 106 nm and 130 nm, respectively. These gratings are also more fabrication tolerant compared to similar designs not using metamaterials.

The performance of integrated silicon photonic devices is sensitive to small structural variations that arise from imperfections in the nanofabrication process. This sensitivity is exacerbated for next-generation devices that require fine feature sizes to push the limits of performance. In this work, we present a deep convolutional neural network model to predict fabrication variations in planar silicon photonic devices and verify their manufacturing feasibility prior to prototyping. Our model is trained on a modest set of scanning electron microscope images of structures that experience dimensional inaccuracies stemming from combined contributions from proximity effects in lithography and loading effects in dry etching. Our model quickly and accurately predicts over/under-etching, corner rounding, filling of narrow channels and holes, and washing away of small features in a photonic device. With this, the expected performance of a device can be predicted through an extra simulation and any necessary design corrections can be made prior to fabrication.

In this paper, we introduce an energy constraint to improve topology-based inverse design. Current methods typically place the constraints solely on the device geometry and require many optimization iterations to converge to a manufacturable solution. In our approach the energy constraint directs the optimization process to solutions that best contain the optical field inside the waveguide core medium, leading to more robust designs with relatively larger minimum feature size. To validate our method, we optimize two components: a mode converter (MC) and a wavelength demultiplexer. In the MC, the energy constraint leads to nearly binarized structures without applying independent binarization stage. In the demultiplexer, it also reduces the appearance of small features. Furthermore, the proposed constraint improves the robustness to fabrication imperfections as shown in demultiplexer design. With energy constraint optimization, the corresponding spectrum shifts under ±10 nm dimensional variations are reduced by 17% to 30%. The proposed constraint is unique in simultaneously taking both geometry and electric field into account, opening the door to new ideas and insights to further improve the computationally intensive topology-based optimization process of nanophotonic devices.