Parallel Dot Products Using Silicon Photonics

This paper proposes a parallel photonic architecture for computing dense dot products, such as those found during deep neural network (DNN) inference, and quantifies the architecture's computation error induced by crosstalk and noise.