Recently, the National Natural Science Foundation of China reported on the progress made by the School of Optoelectronics at University of Shanghai for Science and Technology in the research of polarization-selective diffraction neural networks under the title "Chinese Scholars Make Progress in Polarization-selective Diffraction Neural Networks" in the "Funded Achievements" column on its homepage. The report is as follows: All-optical information processing has advantages such as fast response speed, low loss and high parallelism, and has important applications in many fields such as image recognition, video analysis and intelligent sensing. As the main carrier of all-optical information processing, the all-optical diffraction neural network realizes the rapid processing of optical signals through interlayer optical diffraction. However, traditional all-optical diffraction neural networks are restricted by their reciprocity characteristics and are unable to achieve the functional switch from unidirectional to bidirectional information processing, making it difficult to balance information security and data sharing. Facing users' diverse information acquisition demands, how to achieve data sharing while ensuring information security is one of the difficult problems that need to be solved in the field of all-optical information processing. Under the National Natural Science Foundation of China (Grant No. Under the support of 61988102,62271320, a team led by Academician Zhuang Songlin, Professor Zhu Yiming and Professor Zang Xiaofei from University of Shanghai for Science and Technology, in collaboration with a team led by Professor Ding Fei from Ningbo Oriental University of Technology (provisional name), proposed an all-optical information processing method combining phase control and polarization rotation to address the aforementioned challenges. By controlling the in-plane rotation direction of the metasurface unit structure, By regulating the phase and polarization state during the light diffraction process, a polarization switchable diffraction neural network architecture can be constructed. This architecture realizes unidirectional information transmission under x-/y- polarization incidence and bidirectional information transmission under 45° linear polarization incidence, forming asymmetric all-optical diffraction processing and solving the problem of functional switching from unidirectional to bidirectional optical signal transmission. Furthermore, the team designed a cascaded double-layer metasurface classification and imaging device and constructed a polarization with 100×100 neurons
