Integrated diffractive deep neural networks for near-infrared and visible wavelengths

Diffractive deep neural networks (DDNNs) are free-space optical image processors that can perform image classifications. Integrated DDNNs can be directly fabricated on detectors with two-photon polymerization (TPP). This publication discusses some of the numerical and fabrication limitations that govern the design of integrated DDNNs for near-infrared (NIR) and visible wavelengths. Numerical optimization and simulation results are presented for an example geometry that can be fabricated with TPP printing.