Study of simulation model of diffractive neural network in liquid crystal optical environments

The optical environment plays an important role in modulation of nanophotonics devices, which potentially would make it possible in development of tunable optical diffractive neural network (ODNN), which would significantly progress the capability in machine learning tasks. However, most of the current research on ODNN is focused on passive nanostructures in atmospheric environments, which only gives a fixed optical response of learning, while limited discussion on the performance has been reported in other tunable spatial environments. In this paper, model of the training and testing effects of ODNN in different optical environments, especially in birefringent nematic liquid crystal based on the Rayleigh-Sommerfeld diffraction is systematically investigated and analyzed. ODNN models are trained in air, water and liquid crystal environments at the visible light wavelength of 532 nm. The results indicate that ODNN is sensitive to changes in the testing environment, and the inference capability of the network degrades as the deviation between the testing environment and training conditions becomes significant. Different learning tasks can be carried out by tuning of the optical environment in device. Since liquid crystal is a widely used electronic material tunable under different external physics conditions, these results provide a starting point to introduce more effective dimensions of optical learning in a tunable ODNN electronic device.