Integrated semiconductor optical amplifiers based photonic cross-connect for deep neural networks

Bin Shi; Nicola Calabretta; Ripalta  Stabile

Integrated semiconductor optical amplifiers based photonic cross-connect for deep neural networks

Bin Shi, Nicola Calabretta, Ripalta Stabile

Research output: Contribution to conference › Abstract › Academic

Abstract

We perform weight addition of four 10 Gbit/s channels employing a photonic integrated indium phosphide (InP) chip based on arrayed waveguide gratings (AWGs) and semiconductor optical amplifiers (SOAs) technology. Continuous weight tuning for positive and negative weight matrix numbers is demonstrated with a maximum reading error of 0.04. The utilized chip is already designed to perform cross-connectivity between two layers populated with 8 neurons each, for the complete integration of one input layer and one hidden layer for photonic multilayer perceptron neural network.

Original language	English
Publication status	Published - 2018
Event	Conference in Cognitive Computing 2018: Merging Concepts with Hardware - Herrenhausen, Hannover, Germany Duration: 18 Dec 2018 → 20 Dec 2018 http://www.cognitive-comp.org/

Conference

Conference	Conference in Cognitive Computing 2018
Country/Territory	Germany
City	Hannover
Period	18/12/18 → 20/12/18
Internet address	http://www.cognitive-comp.org/

Cite this

@conference{1829557ce8014a9db467581f1e94ccff,

title = "Integrated semiconductor optical amplifiers based photonic cross-connect for deep neural networks",

abstract = "We perform weight addition of four 10 Gbit/s channels employing a photonic integrated indium phosphide (InP) chip based on arrayed waveguide gratings (AWGs) and semiconductor optical amplifiers (SOAs) technology. Continuous weight tuning for positive and negative weight matrix numbers is demonstrated with a maximum reading error of 0.04. The utilized chip is already designed to perform cross-connectivity between two layers populated with 8 neurons each, for the complete integration of one input layer and one hidden layer for photonic multilayer perceptron neural network.",

author = "Bin Shi and Nicola Calabretta and Ripalta Stabile",

year = "2018",

language = "English",

note = "Conference in Cognitive Computing 2018 : Merging Concepts with Hardware ; Conference date: 18-12-2018 Through 20-12-2018",

url = "http://www.cognitive-comp.org/",

}

TY - CONF

T1 - Integrated semiconductor optical amplifiers based photonic cross-connect for deep neural networks

AU - Shi, Bin

AU - Calabretta, Nicola

AU - Stabile, Ripalta

PY - 2018

Y1 - 2018

N2 - We perform weight addition of four 10 Gbit/s channels employing a photonic integrated indium phosphide (InP) chip based on arrayed waveguide gratings (AWGs) and semiconductor optical amplifiers (SOAs) technology. Continuous weight tuning for positive and negative weight matrix numbers is demonstrated with a maximum reading error of 0.04. The utilized chip is already designed to perform cross-connectivity between two layers populated with 8 neurons each, for the complete integration of one input layer and one hidden layer for photonic multilayer perceptron neural network.

AB - We perform weight addition of four 10 Gbit/s channels employing a photonic integrated indium phosphide (InP) chip based on arrayed waveguide gratings (AWGs) and semiconductor optical amplifiers (SOAs) technology. Continuous weight tuning for positive and negative weight matrix numbers is demonstrated with a maximum reading error of 0.04. The utilized chip is already designed to perform cross-connectivity between two layers populated with 8 neurons each, for the complete integration of one input layer and one hidden layer for photonic multilayer perceptron neural network.

M3 - Abstract

T2 - Conference in Cognitive Computing 2018

Y2 - 18 December 2018 through 20 December 2018

ER -

Integrated semiconductor optical amplifiers based photonic cross-connect for deep neural networks

Abstract

Conference

Fingerprint

Cite this