Free-space and solid-matrix-media diffraction neural network masks made by two-photon lithography

Neural networks are powerful tools for solving many modern problems. One of the options for the optical implementation of a neural network is a diffraction neural network, which consists of one or several layers of different-sized pixels on which radiation diffracts. The pixel parameters are tightly bound with the desired radiation wavelength. In this work, we printed masks for diffraction neural networks for the optical range using two-photon laser lithography. Applying coordinate stabilization approach and preserving temperature and humidity allowed to print pixels with up to 10 nm height difference and 2.3 nm average surface roughness.