Deep Learning Based Inverse Design of Nanoscale Optical Bandpass Filter for Sub-THz 6G Network

Abstract

In this paper, the novel deep learning-based nano scale optical filter is designed with narrow bandwidth for 6G network and Dense Wavelength Division Multiplexing (DWDM) systems. The hybrid Long Short-Term Memory Neural Network (LSTM-NN)-transformer based deep learning algorithm is implemented to accurately predict the structural parameter of the optical bandpass filter. The inverse design approach-based hybrid deep learning model is designed to improve the performance of the optical bandpass filter. The photonic filter performance parameters are numerically analyzed by Finite Difference Time Domain (FDTD) method. The proposed hybrid model is designed with very low mean square error of 5.4207 × 10⁻⁸ and less computation time of 834.81 seconds. The presented photonics platform is designed with narrow bandwidth of 1.12 THz and footprint is very compact as about 134 μm². Therefore, the proposed optical filter is highly suitable for photonic integrated circuits and lightwave communication systems.