Modeling and Design of FTJs as Multi-Level Low Energy Memristors for Neuromorphic Computing

Fontanini, Riccardo; Segatto, M.; Massarotto, Marco; Specogna, Ruben; Driussi, F.; Loghi, Mirko; Esseni, D.

doi:10.1109/jeds.2021.3120200

Cited by 21 publications

(18 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this work, the capture rates were attributed to tunnelling from and to the electrodes, and the tunnelling transmission described according to a WKB approximation that involves the tunnelling effective mass in the two dielectrics m D , m F , and the area and energy cross sections σ T [m 2 ], σ E [eV] (see also Figure 3). More details about the trapping and tunnelling models may be found in [9], [26] reporting also a good agreement with the polarization versus voltage curves in FTJs with an MFDM structure and a thin dielectric layers. The values of m D , m F , σ T , σ E used in this work are reported in Table 1.…”

Section: A Modelling Framework and Validationsupporting

confidence: 53%

Limitations to Electrical Probing of Spontaneous Polarization in Ferroelectric-Dielectric Heterostructures

Segatto

Fontanini

Driussi

et al. 2022

IEEE J. Electron Devices Soc.

Self Cite

View full text Add to dashboard Cite

An accurate estimate of the ferroelectric polarization in ferroelectric-dielectric stacks is important from a materials science perspective, and it is also crucial for the development of ferroelectric based electron devices. This paper revisits the theory and application of the PUND technique in Metal-Ferroelectric-Dielectric-Metal (MFDM) structures by using analytical derivations and numerical simulations. In an MFDM structure the results of the PUND technique may largely differ from the polarization actually switched in the stack, which in turn is different from the remnant polarization of the underlying ferroelectric. The main hindrances that prevent PUND measurements from providing a good estimate of the polarization switching in MFDM stacks are thus discussed. The inspection of the involved physical quantities, not always accessible in experiments, provides a useful insight about the main sources of the errors in the PUND technique, and clarifies the delicate interplay between the depolarization field and the charge injection and trapping in MFDM stacks with a thin dielectric layer.

show abstract

Section: A Modelling Framework and Validationsupporting

confidence: 53%

Limitations to Electrical Probing of Spontaneous Polarization in Ferroelectric-Dielectric Heterostructures

Segatto

Fontanini

Driussi

et al. 2022

IEEE J. Electron Devices Soc.

Self Cite

View full text Add to dashboard Cite

show abstract

“…1 shows the resistive memory model for a memristor, in which the resistance of the model is coupled with the voltage across it. Interestingly, there have been previous papers which have supported the positive impact of advancement in resistive memory on deep learning, particularly computer vision further proposing the impact in medical and sport sciences as well [5][6][7][8][9][10][11][12][13][14][15]. In this work, we perform dual ion beam sputtering and also fabricate a metal oxide memristor consisting of the format: Al/ZnO/Al.…”

Section: Introductionmentioning

confidence: 99%

Modelling Simulation of Resistive RAM devices for Analog and Digital Paradigms: An Experimental Review

Jianyi¹

2022

Preprint

View full text Add to dashboard Cite

Over the years, with the introduction of non-volatile memory devices such as memristors and other CMOS relevant devices, the ability to manufacture hardware which can emulate the properties of a biological synapse has become a realistic possibility. As a result, this has lead to an improvement in the corresponding computer vision and the machine learning applications from a software perspective which can support the advancement in resistive memory technology such as using memristors for neuromorphic applications. However, the increasing complexity of deep learning models has increased the demand for more sophisticated hardware architecture which enables the full potential of such complex networks on devices with resource constraints. In this work, we perform dual ion beam sputtering and also fabricate a metal oxide memristor consisting of the format: Al/ZnO/Al. We found that our memristor displayed characteristics such as bipolar resistive switching characteristics. Moreover, upon investigation, we observed that a partially amorphous thin film of ZnO and the suitable number if vacancies of oxygen help greatly in inducing the memristor to have reliable and stable behaviour in the memory cell regions. Furthermore, upon further investigation, we found that the oxide ions in the lattice can play a significant role in changing in resistance of the memristor. The endurance of the memristor had good retention and reliable voltage difference for over 300 cycles.

show abstract

“…This spurred an interesting line of research on using the concept of spike-timing-dependent-plasticity (STDP) to program the resistance of the memristor [1][2][3][4][5][6][7][8]. This enabled several methods to use the properties of a memristor to emulate an artificial biological synapse.…”

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Designing Efficient Memristors for resource-constrained inference on pruned Deep Neural Networks

Huang¹

2022

Preprint

View full text Add to dashboard Cite

The increasing complexity of deep learning models has increased the demand for more sophisticated hardware architecture which enables the full potential of such complex networks on devices with resource constraints. As a result, this has lead to an improvement in the corresponding computer vision and the machine learning applications from a software perspective which can support the advancement in resistive memory technology such as using memristors for neuromorphic applications. However, the increasing complexity of deep learning models has increased the demand for more sophisticated hardware architecture which enables the full potential of such complex networks on devices with resource constraints. In this work, we perform dual ion beam sputtering and also fabricate a metal oxide memristor consisting of the format: Al/ZnO/Al. We found that our memristor displayed characteristics such as bipolar resistive switching characteristics. Moreover, upon investigation, we observed that a partially amorphous thin film of ZnO and the suitable number if vacancies of oxygen help greatly in inducing the memristor to have reliable and stable behaviour in the memory cell regions. Furthermore, upon further investigation, we found that the oxide ions in the lattice can play a significant role in changing in resistance of the memristor. The endurance of the memristor had good retention and reliable voltage difference for over 300 cycles.

show abstract

Modeling and Design of FTJs as Multi-Level Low Energy Memristors for Neuromorphic Computing

Cited by 21 publications

References 24 publications

Limitations to Electrical Probing of Spontaneous Polarization in Ferroelectric-Dielectric Heterostructures

Limitations to Electrical Probing of Spontaneous Polarization in Ferroelectric-Dielectric Heterostructures

Modelling Simulation of Resistive RAM devices for Analog and Digital Paradigms: An Experimental Review

Designing Efficient Memristors for resource-constrained inference on pruned Deep Neural Networks

Contact Info

Product

Resources

About