The advent of Artificial Intelligence (AI) and big data era brought an unprecedented (and ever growing) need for fast and energy efficient computation that cannot be obtained with the classical von Neumann computing architecture. This paved the way to new technologies that try to mimic the human brain to leverage its energy efficient and distributed computing. Nonvolatile memory technologies seem to be the ideal solution for the hardware implementation of artificial neurons and synapses of Neural Network (NN) architectures and have been extensively investigated in the last years. However, they often suffer from limited linearity and symmetry, poor retention, and high variability, thus requiring significant advancements for their mass adoption in NNs. One of the cornerstones on which the required efforts must be based is certainly represented by device simulations, which can be effectively used to achieve a full understanding of the physics governing the device, which in turn is a pre-requisite for their design, optimization and successful exploitation in neuromorphic applications. In this scenario, we focus on Transition Metal Oxide (TMO)-based RRAM, Ferroelectric Tunnel Junctions (FTJ) and 3D-NAND Charge Trap devices and use simulations to address key issues related to neuromorphic operation (linearity and asymmetry) and reliability (variability and retention).
Reliability of Non-Volatile Memory Devices for Neuromorphic Applications: A Modeling Perspective (Invited) / Padovani, A.; Pesic, M.; Nardi, F.; Milo, V.; Larcher, L.; Kumar, M. A.; Baten, M. Z.. - 2022-:(2022), pp. 3C41-3C410. (Intervento presentato al convegno 2022 IEEE International Reliability Physics Symposium, IRPS 2022 tenutosi a Dallas, TX, USA nel 2022) [10.1109/IRPS48227.2022.9764451].
Reliability of Non-Volatile Memory Devices for Neuromorphic Applications: A Modeling Perspective (Invited)
Padovani A.
;Larcher L.;
2022
Abstract
The advent of Artificial Intelligence (AI) and big data era brought an unprecedented (and ever growing) need for fast and energy efficient computation that cannot be obtained with the classical von Neumann computing architecture. This paved the way to new technologies that try to mimic the human brain to leverage its energy efficient and distributed computing. Nonvolatile memory technologies seem to be the ideal solution for the hardware implementation of artificial neurons and synapses of Neural Network (NN) architectures and have been extensively investigated in the last years. However, they often suffer from limited linearity and symmetry, poor retention, and high variability, thus requiring significant advancements for their mass adoption in NNs. One of the cornerstones on which the required efforts must be based is certainly represented by device simulations, which can be effectively used to achieve a full understanding of the physics governing the device, which in turn is a pre-requisite for their design, optimization and successful exploitation in neuromorphic applications. In this scenario, we focus on Transition Metal Oxide (TMO)-based RRAM, Ferroelectric Tunnel Junctions (FTJ) and 3D-NAND Charge Trap devices and use simulations to address key issues related to neuromorphic operation (linearity and asymmetry) and reliability (variability and retention).File | Dimensione | Formato | |
---|---|---|---|
(A. Padovani - IRPS 2022) Reliability of Non-Volatile Memory Devices for Neuromorphic Applications - A Modeling Perspective (Invited).pdf
Accesso riservato
Tipologia:
VOR - Versione pubblicata dall'editore
Dimensione
2.22 MB
Formato
Adobe PDF
|
2.22 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
Pubblicazioni consigliate
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris