Scalable intensity-based photonic matrix-vector multiplication processor using single-wavelength time-division-multiplexed signals

TL;DR

This paper presents a scalable, single-wavelength intensity-based photonic matrix-vector multiplication processor fabricated on SOI, demonstrating effective convolution operations in CNNs with high accuracy, simplifying optical processing.

Contribution

It introduces a novel TDM-based photonic MVM processor using only intensity modulation, avoiding complex detection methods, and demonstrates its application in CNN digit recognition.

Findings

Successfully fabricated a 32-channel SOI photonic processor.

Achieved 93.47% accuracy in handwritten digit recognition.

Demonstrated scalable convolution operations in neural networks.

Abstract

Photonic integrated circuits provide a compact platform for ultrafast and energy-efficient matrix-vector multiplications (MVMs) in the optical domain. Recently, schemes based on time-division multiplexing (TDM) have been proposed as scalable approaches for realizing large-scale photonic MVM processors. However, existing demonstrations rely on coherent detection or multiple wavelengths, both of which complicate their operations. In this work, we demonstrate a scalable TDM-based photonic MVM processor that uses only single-wavelength intensity-modulated optical signals, thereby avoiding coherent detection and enabling simplified operations. A 32-channel processor is fabricated on a Si-on-insulator (SOI) platform and used to experimentally perform convolution operations in a convolutional neural network (CNN) for handwritten digit recognition, achieving a classification accuracy of 93.47% for 1500 images.