We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.ET

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Emerging Technologies

Title: A Single-Cycle MLP Classifier Using Analog MRAM-based Neurons and Synapses

Authors: Ramtin Zand
Abstract: In this paper, spin-orbit torque (SOT) magnetoresistive random-access memory (MRAM) devices are leveraged to realize sigmoidal neurons and binarized synapses for a single-cycle analog in-memory computing (IMC) architecture. First, an analog SOT-MRAM-based neuron bitcell is proposed which achieves a 12x reduction in power-area-product compared to the previous most power- and area-efficient analog sigmoidal neuron design. Next, proposed neuron and synapse bit cells are used within memory subarrays to form an analog IMC-based multilayer perceptron (MLP) architecture for the MNIST pattern recognition application. The architecture-level results exhibit that our analog IMC architecture achieves at least two and four orders of magnitude performance improvement compared to a mixed-signal analog/digital IMC architecture and a digital GPU implementation, respectively while realizing a comparable classification accuracy.
Comments: arXiv admin note: substantial text overlap with arXiv:2006.01238
Subjects: Emerging Technologies (cs.ET); Hardware Architecture (cs.AR); Machine Learning (cs.LG)
Cite as: arXiv:2012.02695 [cs.ET]
  (or arXiv:2012.02695v1 [cs.ET] for this version)

Submission history

From: Ramtin Zand [view email]
[v1] Fri, 4 Dec 2020 16:04:32 GMT (2065kb,D)

Link back to: arXiv, form interface, contact.