Automated design of freeform optical systems by reverse ray-tracing and AI

Freeform optics offer several potential advantages over spheric and aspheric systems. These range from improved optical quality to overall compactness of the system. However, their adoption has been slow due to the complexity of the design and fabrication process. Here, methods that facilitate the design of freeform optical systems (including freeform arrays) are presented: deep neural network (DNN) based selection of the initial layout, topological optimization of freeform segment layouts, and reverse ray-tracing based construction of freeform surfaces. By combining these methods into a streamlined automatic design process, high performance optical systems can be quickly generated from simple specification sheets (e.g. field-of-view, numerical aperture and focal length of target system). When combined with the growing availability of third-party manufacturing of freeform optics using fast-tool servo equipped ultra-precision lathes, this simple design toolchain is attractive for adoption in the education and hobbyist sector.