High optical storage density using three-dimensional hybrid nanostructures based on machine learning

The quality of optical storage is determined by several factors, especially diffraction limit, storage media, record-ing dimensions and readout schemes. To increase the storage density, more progressive strategies reaching high storage density need be explored, including excellent storage media, multiple recording dimensions and robust readout schemes. In this work, the three-dimensional hybrid nanostructures are proposed, which are composed by Si and Si3N4. It can be fabricated with the three dimensional (3D) lithography technology, which is able to encode high density data information within multi-dimensions. To obtain robust information readout, spectra with a broad wavelength range and with a large angle range are combined for a high sensitivity. Meanwhile an artificial neural network is used to translate the spectra to the data sequences. For a 4-unit nanostructure, the 27 data sequences with 20% spectral noise, a 100% readout accuracy can be achieved. High readout speed is also demanded in optical storage. Therefore, we reduced the number of the selected wavelengths and angles in the spectra. It is demonstrated that when the number reduces to 1/50 of the original number, the reading accuracy is still 99.7% compared to the full range. Furthermore, the possibility of achieving higher storage densities is investigated for a further study.