2D photonic crystal optical demultiplexer with simplified structure using deep neural networks

This paper presents the design and analysis of an all-optical 1×2 digital demultiplexer based on a 2D photonic crystal structure composed of silicon rods in an air background. The device is implemented on a square lattice and operates at the telecommunication wavelength of 1.55 µm, which lies inside the transverse magnetic photonic band gap and therefore allows guided wave propagation through engineered line defects. The proposed demultiplexer uses only linear silicon rods and a very small number of defect rods, namely one defect at the input of the data waveguide and three additional defects for performance optimization. This is in contrast to the only two previously reported photonic crystal demultiplexers, which all require nonlinear rods and a large number of closely spaced defects, making fabrication very challenging. The finite difference time domain method is used to verify correct logical behavior for all four input states and to extract key performance parameters. The structure achieves a response time of 0.2 ps and a bit rate of 5 Tbit/s, which is higher than the reported values in earlier works. In addition, a deep neural network (DNN) with three hidden layers is employed to search the design space and identify an optimized photonic crystal configuration, reducing the number of required full-wave simulations and providing a systematic way to obtain the best structure for the proposed demultiplexer.