In this study, we report an innovative multi-parameter artificial neural network (ANN) based optimization approach for designing InP-based capacitive loading traveling wave Mach-Zehnder modulators (CL-MZMs). Our ANN-based heuristic algorithm optimization method surpasses traditional manual optical device design in efficiently searching for the optimal solution, based on user-defined figure of merit (FOM) in a large multi-parameter design space, while also providing statistical data-based insight into the underlying complex device physics involved. We achieved an optimized 1 mm InP MZM design, with an anticipated 112 GHz 3-dB electro-optic bandwidth and 5.8 V half-wave voltage, making it a promising candidate for next-generation data center high-speed optical link applications at 400 Gb/s and beyond.
Read full abstract