3D computer-generated holography by non-convex optimization

3D computer-generated holography uses a digital phase mask to shape the wavefront of a laser beam into a user-specified 3D intensity pattern. Algorithms take the target 3D intensity as input and compute the hologram that generates it. However, arbitrary patterns are generally infeasible, so solutions are approximate and often sub-optimal. Here, we propose a new non-convex optimization algorithm that computes holograms by minimizing a custom cost function that is tailored to particular applications (e.g., lithography, neural photostimulation) or leverages additional information like sample shape and nonlinearity. Our method is robust and accurate, and it out-performs existing algorithms.