Improved Synaptic Characteristics in Bilayer Memristor by Post-Oxide Deposition Annealing for Pattern Recognition

Dayanand Kumar; Yi-Rong Huang; Pratibha Pal; Aftab Saleem; Amit Singh; Hoonkyung Lee; Yeong-Her Wang; Teung-Yuen Tseng

doi:10.1109/VLSI-TSA54299.2022.9771007

Improved Synaptic Characteristics in Bilayer Memristor by Post-Oxide Deposition Annealing for Pattern Recognition

Dayanand Kumar, Yi-Rong Huang, Pratibha Pal, Aftab Saleem, Amit Singh, Hoonkyung Lee, Yeong-Her Wang, Teung-Yuen Tseng

Academia Sinica

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

Abstract

We present highly stable bilayer HfO2/TiO2 memristive synapse for neuromorphic computing applications. The memristive synapse shows repetitive 220 Long-term potentiation and depression cycles without any breakdown with total number of 220k pulses. The nonlinear values of potentiation and depression are 2.52 and -2.63, respectively with 1000 conductance pulses (500 pulses of potentiation and 500 pulses of depression). The experimental data of potentiation and depression was used to train HNN for pattern recognition of 28×28 pixels comprising 784 synapses. In 16 iterations, the HNN can be successfully trained to identify the input image with a training accuracy of about 97%. Moreover, the device has good retention (10^4 s) at 120 °C. These excellent synaptic characteristics of annealed device allows it for artificial intelligence systems in near future....

Original language	Undefined
Title of host publication	2022 International Symposium on VLSI Technology
DOIs	https://doi.org/10.1109/VLSI-TSA54299.2022.9771007
Publication status	Published - May 2022
Externally published	Yes

Access to Document

10.1109/VLSI-TSA54299.2022.9771007

Cite this

@inproceedings{ab3d845ab9af480bbbed6439afd0787d,

title = "Improved Synaptic Characteristics in Bilayer Memristor by Post-Oxide Deposition Annealing for Pattern Recognition",

abstract = "We present highly stable bilayer HfO2/TiO2 memristive synapse for neuromorphic computing applications. The memristive synapse shows repetitive 220 Long-term potentiation and depression cycles without any breakdown with total number of 220k pulses. The nonlinear values of potentiation and depression are 2.52 and -2.63, respectively with 1000 conductance pulses (500 pulses of potentiation and 500 pulses of depression). The experimental data of potentiation and depression was used to train HNN for pattern recognition of 28×28 pixels comprising 784 synapses. In 16 iterations, the HNN can be successfully trained to identify the input image with a training accuracy of about 97%. Moreover, the device has good retention (10^4 s) at 120 °C. These excellent synaptic characteristics of annealed device allows it for artificial intelligence systems in near future....",

author = "Dayanand Kumar and Yi-Rong Huang and Pratibha Pal and Aftab Saleem and Amit Singh and Hoonkyung Lee and Yeong-Her Wang and Teung-Yuen Tseng",

year = "2022",

month = may,

doi = "10.1109/VLSI-TSA54299.2022.9771007",

language = "Undefined",

booktitle = "2022 International Symposium on VLSI Technology",

}

TY - GEN

T1 - Improved Synaptic Characteristics in Bilayer Memristor by Post-Oxide Deposition Annealing for Pattern Recognition

AU - Kumar, Dayanand

AU - Huang, Yi-Rong

AU - Pal, Pratibha

AU - Saleem, Aftab

AU - Singh, Amit

AU - Lee, Hoonkyung

AU - Wang, Yeong-Her

AU - Tseng, Teung-Yuen

PY - 2022/5

Y1 - 2022/5

N2 - We present highly stable bilayer HfO2/TiO2 memristive synapse for neuromorphic computing applications. The memristive synapse shows repetitive 220 Long-term potentiation and depression cycles without any breakdown with total number of 220k pulses. The nonlinear values of potentiation and depression are 2.52 and -2.63, respectively with 1000 conductance pulses (500 pulses of potentiation and 500 pulses of depression). The experimental data of potentiation and depression was used to train HNN for pattern recognition of 28×28 pixels comprising 784 synapses. In 16 iterations, the HNN can be successfully trained to identify the input image with a training accuracy of about 97%. Moreover, the device has good retention (10^4 s) at 120 °C. These excellent synaptic characteristics of annealed device allows it for artificial intelligence systems in near future....

AB - We present highly stable bilayer HfO2/TiO2 memristive synapse for neuromorphic computing applications. The memristive synapse shows repetitive 220 Long-term potentiation and depression cycles without any breakdown with total number of 220k pulses. The nonlinear values of potentiation and depression are 2.52 and -2.63, respectively with 1000 conductance pulses (500 pulses of potentiation and 500 pulses of depression). The experimental data of potentiation and depression was used to train HNN for pattern recognition of 28×28 pixels comprising 784 synapses. In 16 iterations, the HNN can be successfully trained to identify the input image with a training accuracy of about 97%. Moreover, the device has good retention (10^4 s) at 120 °C. These excellent synaptic characteristics of annealed device allows it for artificial intelligence systems in near future....

U2 - 10.1109/VLSI-TSA54299.2022.9771007

DO - 10.1109/VLSI-TSA54299.2022.9771007

M3 - Conference contribution

BT - 2022 International Symposium on VLSI Technology

ER -