In-situ optical tomographic reconstruction during 3D laser microprinting using deep learning (Invited Paper)

3D laser-printed microstructures often differ from the intended models due to various mechanisms, such as dose accumulation, shrinkage, or unintended printing below the substrate. So far, the deviations between the intended model and the ex-situ characterization result had to be compensated iteratively, leading to a tedious feedback loop. Here, we present a novel deep learning-driven in-situ tomographic reconstruction technique based on stacks of widefield optical intensity images taken during the printing process. A deep neural network is trained to reconstruct specimens by simulated optical intensity images. The reconstruction before development during the printing process itself can drastically accelerate material design and characterization.