Stochastic weight updates in phase-change memory-based synapses and their influence on artificial neural networks
Lists
Only you see this
Abstract
Artificial neural networks (ANN) have become a<br> powerful tool for machine learning. Resistive memory devices can<br> be used for the realization of a non-von Neumann computational<br> platform for ANN training in an area-efficient way. For instance,<br> the conductance values of phase-change memory (PCM) devices<br> can be used to represent synaptic weights and can be updated<br> in-situ according to learning rules. However, non-ideal device<br> characteristics pose challenges to reach competitive classification<br> accuracies. In this paper, we investigate the impact of granularity<br> and stochasticity associated with the conductance changes on<br> ANN performance. Using a PCM prototype chip fabricated in<br> the 90nm technology node, we present a detailed experimental<br> characterization of the conductance changes. Simulations are<br> done in order to quantify the effect of the experimentally<br> observed conductance change granularity and stochasticity on<br> classification accuracies in a fully connected ANN trained with<br> backpropagation.
BibTeX
@inproceedings{boybat2017stochastic,
author = {Irem Boybat and Manuel Le Gallo and Timoleon Moraitis and Yusuf Leblebici and Abu Sebastian and Evangelos Eleftheriou},
title = {{Stochastic weight updates in phase-change memory-based synapses and their influence on artificial neural networks}},
booktitle = {Zenodo},
year = {2017},
doi = {10.5281/zenodo.5301477},
}Loading related content...