Delocalized and Single-Chip Approaches to Photonic Machine Learning

Ryan Hamerly,
Massachusetts Institute of Technology & NTT Research

With deep-learning workloads growing and Moore's Law running out of steam, photonics has attracted renewed interest as a computation platform.  However, harnessing the intrinsic advantages of photons--their bandwidth, energy efficiency, and transmissibility--is very challenging in realistic systems.  In this talk, I report two developments in our group aimed at realizing practical photonic deep learning.  I’ll review the scaling challenges inherent in "deep" interferometer circuits (e.g. beamsplitter meshes) and show that fabrication errors limit the circuit size in realistic cases. I’ll also report on our experimental realization of Netcast, a photonic-enabled edge computing scheme that utilizes WDM, integration detection, and optical weight delivery to perform datacenter-scale DNN inference on SWaP-constrained edge devices.  We demonstrate Netcast over an 86-km deployed fiber using 3 THz of optical bandwidth and show minimal loss of accuracy.

Host: Peter McMahon

About the speaker: Ryan Hamerly was born in San Antonio, Texas in 1988. In 2016 he received a Ph.D. degree in applied physics from Stanford University, California, for work with Prof. Hideo Mabuchi on quantum control, nanophotonics, and nonlinear optics. In 2017 he was at the National Institute of Informatics, Tokyo, Japan, working with Prof. Yoshihisa Yamamoto on quantum annealing and optical computing concepts. He is currently a Senior Scientist at NTT PHI Laboratories and a visiting scientist at MIT, Cambridge, Massachusetts, with Prof. Dirk Englund.