Convolutional neural networks for projection multi-photon 3D printing

Micro and nanoscale additive manufacturing using projection multi-photon lithography has the potential to print 3D structures at high speeds. Optimizing parameters for precise 2D layer printing by trial and error requires time-consuming and costly methods. This study introduces a convolutional neural network machine learning scheme to optimize printing using a fast and inexpensive data collection method. By training autoencoders with input patterns and optical microscope images, we can visualize how printed layers would look and explore input layer pattern generation from an inverse model, significantly reducing time and cost in achieving precise micro-nanoscale 3D printed structures.