Adaptive beam forming across temperature variation in optical phased array enabled with deep neural network

Abstract

Integrated optical phased arrays (OPA) require calibration to account for mismatches amongst the channels. Furthermore, beams emitted from an OPA tend to distort when the chip’s temperature changes. We propose to utilize a deep neural network (DNN) to adaptively control the phase modulator voltages of the OPA and create a desired beam pattern in the presence of process mismatches and temperature changes. As a proof of concept, adaptive beam forming was demonstrated with an integrated 128-channel OPA realized in a commercial foundry silicon photonics (SiP) process. Beam forming within 50° field of view (FoV) is demonstrated, while accuracy of 0.025° is achieved when the beam is swept in 0.1° step at a fixed temperature. The DNN is also used to create beams with multiple peaks at desired spatial angles. The DNN is shown to properly adjust the phase modulator voltages to keep the beam nearly intact as temperature changes within 20°C range.