Efficient learning and crossbar operations with atomically-thin 2-D material compound synapses

Ivan Sanchez Esqueda; Huan Zhao; Han Wang

doi:10.1063/1.5042468

Efficient learning and crossbar operations with atomically-thin 2-D material compound synapses

Ivan Sanchez Esqueda, Huan Zhao, Han Wang

Research output: Contribution to journal › Article › peer-review

13 Scopus citations

Abstract

Accurate and efficient synaptic weight programming and vector-matrix multiplication are demonstrated using compound synapses constructed with ultralow power binary memristive devices having oxidized atomically thin two-dimensional hexagonal boron nitride (BNO_x) filament formation layers. Experimental data of the resistive-switching current-voltage characteristics of BNO_x memristors are used to formulate variation-aware models that enable statistically analyzing the trade-off between efficiency and accuracy as a function of the synaptic resolution (i.e., levels of synaptic weight programming). Results are compared with commonly reported oxide-based memristors indicating orders of magnitude (i.e., ∼10⁵) improvements in power efficiency and ∼2-5× improvements in accuracy.

Original language	English
Article number	152133
Journal	Journal of Applied Physics
Volume	124
Issue number	15
DOIs	https://doi.org/10.1063/1.5042468
State	Published - Oct 21 2018
Externally published	Yes

Funding

The authors acknowledge the support from the Army Research Office (Grant no. W911NF-16-1-0435).

Funders	Funder number
Army Research Office	W911NF-16-1-0435

Access to Document

10.1063/1.5042468

Cite this

@article{b7efa0f4c889434390df2be421de43f5,

title = "Efficient learning and crossbar operations with atomically-thin 2-D material compound synapses",

abstract = "Accurate and efficient synaptic weight programming and vector-matrix multiplication are demonstrated using compound synapses constructed with ultralow power binary memristive devices having oxidized atomically thin two-dimensional hexagonal boron nitride (BNOx) filament formation layers. Experimental data of the resistive-switching current-voltage characteristics of BNOx memristors are used to formulate variation-aware models that enable statistically analyzing the trade-off between efficiency and accuracy as a function of the synaptic resolution (i.e., levels of synaptic weight programming). Results are compared with commonly reported oxide-based memristors indicating orders of magnitude (i.e., ∼105) improvements in power efficiency and ∼2-5× improvements in accuracy.",

author = "{Sanchez Esqueda}, Ivan and Huan Zhao and Han Wang",

note = "Publisher Copyright: {\textcopyright} 2018 Author(s).",

year = "2018",

month = oct,

day = "21",

doi = "10.1063/1.5042468",

language = "English",

volume = "124",

journal = "Journal of Applied Physics",

issn = "0021-8979",

publisher = "American Institute of Physics",

number = "15",

}

TY - JOUR

T1 - Efficient learning and crossbar operations with atomically-thin 2-D material compound synapses

AU - Sanchez Esqueda, Ivan

AU - Zhao, Huan

AU - Wang, Han

PY - 2018/10/21

Y1 - 2018/10/21

N2 - Accurate and efficient synaptic weight programming and vector-matrix multiplication are demonstrated using compound synapses constructed with ultralow power binary memristive devices having oxidized atomically thin two-dimensional hexagonal boron nitride (BNOx) filament formation layers. Experimental data of the resistive-switching current-voltage characteristics of BNOx memristors are used to formulate variation-aware models that enable statistically analyzing the trade-off between efficiency and accuracy as a function of the synaptic resolution (i.e., levels of synaptic weight programming). Results are compared with commonly reported oxide-based memristors indicating orders of magnitude (i.e., ∼105) improvements in power efficiency and ∼2-5× improvements in accuracy.

AB - Accurate and efficient synaptic weight programming and vector-matrix multiplication are demonstrated using compound synapses constructed with ultralow power binary memristive devices having oxidized atomically thin two-dimensional hexagonal boron nitride (BNOx) filament formation layers. Experimental data of the resistive-switching current-voltage characteristics of BNOx memristors are used to formulate variation-aware models that enable statistically analyzing the trade-off between efficiency and accuracy as a function of the synaptic resolution (i.e., levels of synaptic weight programming). Results are compared with commonly reported oxide-based memristors indicating orders of magnitude (i.e., ∼105) improvements in power efficiency and ∼2-5× improvements in accuracy.

UR - http://www.scopus.com/inward/record.url?scp=85055122061&partnerID=8YFLogxK

U2 - 10.1063/1.5042468

DO - 10.1063/1.5042468

M3 - Article

AN - SCOPUS:85055122061

SN - 0021-8979

VL - 124

JO - Journal of Applied Physics

JF - Journal of Applied Physics

IS - 15

M1 - 152133

ER -

Efficient learning and crossbar operations with atomically-thin 2-D material compound synapses

Abstract

Funding

Access to Document

Other files and links

Fingerprint

Cite this