Inverse optimization design of terahertz topological waveguides towards on-chip communication

This study proposes an inverse design framework for terahertz topological waveguides to achieve optimal on-chip communication performances. The topological waveguides are based on valley photonic crystals, which support two low-loss transmission bands (0.25 - 0.29 THz and 0.29 - 0.34 THz) within the topological bandgap. We train a deep neural network (DNN) to achieve high-precision prediction from photonic structural parameters to waveguide performances (including topological bandgap, group velocity dispersion, and transmission loss). After 30,000 epochs, the DNN prediction error is down to 10^-6.5. Subsequently, combining the DNN of topological waveguides with a terahertz on-chip communication link, a direct mapping relationship (i.e., a forward design flow) between the waveguide structure and the communication performance [including the bit error rate (BER) and transmission bandwidth] is established. Furthermore, the particle swarm optimization is applied to the forward design model to inverse optimization the topological waveguide structure, aiming to maximize the communication bandwidth while meeting the forward error correction threshold (BER below 10^-3). Optimization results show that at frequencies around 0.28 THz and 0.31 THz, the topological waveguide supports two low-BER 3.33-Gbps terahertz on-chip communication windows, with bandwidths of 14.81 GHz and 13.67 GHz, respectively. This research realizes the "on-demand" inverse design of microscopic waveguide structures based on macroscopic communication performance requirements, providing an efficient and convenient approach for intelligent optimization of terahertz on-chip communication waveguides.