Digital Twin of All Optical AND Logic Gate Based on Deep Learning

As a fundamental component of optical signal processing, the normal operation of an all optical logic gate is the foundation for ensuring the operation of all optical networks. Using deep learning methods to model the all optical logic gate can effectively avoid complex physical analysis processes, and trained networks can predict outputs based on inputs. In this paper, we use deep learning bidirectional LSTM algorithm with attention mechanism to model an all optical logic AND gate implemented using SOA-MZI. To ensure that SOA operates within the normal injection current operating range, a training set with injection current as a feature is used to train the bidirectional LSTM algorithm with attention mechanism. The results demonstrate that this algorithm can effectively model all optical logic AND gates by comparing the difference between simulation output and digital twin output. Meanwhile, prior estimation of the normal operating range of the injected current can be performed.