Table of Contents
Fetching ...

Enabling Scalable Photonic Tensor Cores with Polarization-Domain Photonic Computing

Amin Shafiee, Linhong Chen, Sudeep Pasricha, Jie Yao, Mahdi Nikdast

Abstract

We present a silicon-photonic tensor core using 2D ferroelectric materials to enable wavelength- and polarization-domain computing. Results, based on experimentally characterized material properties, show up to 83% improvement in computation accuracy compared to coherent networks.

Enabling Scalable Photonic Tensor Cores with Polarization-Domain Photonic Computing

Abstract

We present a silicon-photonic tensor core using 2D ferroelectric materials to enable wavelength- and polarization-domain computing. Results, based on experimentally characterized material properties, show up to 83% improvement in computation accuracy compared to coherent networks.

Paper Structure

This paper contains 3 sections, 3 figures.

Figures (3)

  • Figure 1: (a) Molecular structure of SnSe in two nonvolatile ferroelectric states of armchair and zigzag. (b) Experimentally characterized (b) refractive index and (c) extiction coefficient of the SnSe in different in-plane crystalline orientations. (d) Microscopic image of the SnSe film from our experiments.
  • Figure 2: (a) Mapping the multiplication results onto Stokes parameters by changing the ellipticity angle of the polarization. (b) Design of the wavelength-multiplexed polarization-domain photonic tensor core.
  • Figure 3: Optical transmission of PBS when (a) TE0 and (b) TM0 source is used. (c) $S_2$ of polarized light interacting with MX versus propagation length. (d) Polarization ellipse and electric field when $\phi=$1 (rad). The design-space exploration of phase shift per unit length for (e) TM polarization and (f) TE polarization. (g) RVD and inferencing accuracy values for WPol-PTC and coherent networks of different sizes. (h) Programming power comparison between WPol-PTC and a MZI-based coherent network both with SnSe phase shifters.