Prediction of Impedance Matching Properties Between SNSPD and Free Space Based on BP Neural Network

Impedance matching between the device and the free space is one of the most important factors that affect the detection efficiency of the superconducting nanowire single-photon detector (SNSPD). Therefore, we present a neural network based method to predict the impedance matching properties between SNSPD and free space in this paper. First, the database of the impedance matching properties at 1310nm for a simple three-layer device structure is established by using the transmission line theory and one million samples are obtained to facilitate the following process. Then, a back propagation neural network (BPNN) with one hidden layer, including one hundred neurons, is trained and tested by the impedance matching database and a root mean square error (RMSE) of 0.138 suggests the acceptable regression capacity of the network. Finally, considering the inherent periodicity of impedance matching properties, two different new datasets with the same matching condition are generated to further illustrate the generalization ability of the BPNN. The RMSE for the dataset inside and outside the range of the impedance matching database are 0.107 and 0.487, respectively.