Memristive devices for deep learning applications

Damien Querlioz, Sabina Spiga, Abu Sebastian, Bipin Rajendran

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

Deep neural networks have become the flagship approach of Artificial Intelligence, and every week, new amazing achievements of such networks are announced. However they come with a challenge: their energy consumption. Deep neural networks running on central or graphical processors can consume thousands times more energy than the brain on similar tasks. Memristive devices are now considered as a fantastic opportunity to reduce the energy consumption of deep learning, and this chapter explains this. First we introduce the general principles of deep neural networks. This allows us to explore to what extent deep neural networks are similar and dissimilar to the brain. In particular we discuss the fundamental reasons for their difference in energy consumption. These considerations made us discuss the opportunities, but also the challenges of implementing deep neural networks with memristive devices, which serves as an introduction for the next two chapters.

Original languageEnglish (US)
Title of host publicationMemristive Devices for Brain-Inspired Computing
Subtitle of host publicationFrom Materials, Devices, and Circuits to Applications - Computational Memory, Deep Learning, and Spiking Neural Networks
PublisherElsevier
Pages313-327
Number of pages15
ISBN (Electronic)9780081027820
DOIs
StatePublished - Jan 1 2020
Externally publishedYes

All Science Journal Classification (ASJC) codes

  • Engineering(all)

Keywords

  • Deep learning
  • Hardware neural networks
  • Memristive devices
  • Memristor
  • Resistive memory

Fingerprint

Dive into the research topics of 'Memristive devices for deep learning applications'. Together they form a unique fingerprint.

Cite this