Litcius/Paper detail

Silicon-integrated coherent neurons with 32GMAC/sec/axon compute line-rates using EAM-based input and weighting cells

George Giamougiannis, Apostolos Tsakyridis, George Mourgias-Alexandris, Miltiadis Moralis‐Pegios, Angelina Totović, George Dabos, Nikos Passalis, Manos Kirtas, Nikos Bamiedakis, Anastasios Tefas, David Lazovsky, Nikos Pleros

202130 citationsDOI

Abstract

We experimentally demonstrate a coherent SiPho neuron that relies on EAM for both on-chip data generation and weighting. A record-high 32GMAC/s/axon compute rate and an accuracy of 95.91% is reported, when the neuron is deployed as a hidden layer of a MNIST classifier neural network.

Topics & Concepts

MNIST databaseWeightingAxonNeuronArtificial neural networkComputer scienceClassifier (UML)Artificial intelligencePattern recognition (psychology)NeurosciencePhysicsBiologyAcousticsNeural Networks and Reservoir ComputingPhotonic and Optical DevicesAdvanced Memory and Neural Computing
Silicon-integrated coherent neurons with 32GMAC/sec/axon compute line-rates using EAM-based input and weighting cells | Litcius