Control of Programmable Optical Processors by Using a Neural-Network Digital Twin
摘要
We propose a neural-network (NN) model to automatically configure programmable optical processors made of meshes of Mach-Zehnder interferometers (MZIs). The NN model predicts the phase shifts of the MZIs such that a MZI mesh implements a given matrix, even in presence of fabrication imperfections and thermal-cross talk. Using a simulated dataset, we validate the NN model and its scalability. Moreover, we show that our approach applies to arbitrary mesh topologies