RRAM-based synapse devices for neuromorphic systems

Moon, Kibong; Lim, Seokjae; Park, J.; Sung, Changhyuck; Oh, Seungyeol; Woo, Jiyong; Lee, J.

doi:10.1039/c8fd00127h

Cited by 192 publications

(152 citation statements)

References 36 publications

Supporting

Mentioning

150

Contrasting

Order By: Relevance

“…Analog incremental current in S4 devices is also beneficial for neuromorphic type applications as compare to the abrupt current change in S2 devices. [ 2,16,25 ]…”

Section: Resultsmentioning

confidence: 99%

“…The cost‐effective, highly‐scalable, high‐density design is capable to show high electrical performance with minimal power consumption. [ 1–4 ] However, reliability and variability of switching are one of the major obstacles to currently existing RRAM technology. [ 5–8 ] In general, it is considered that filament formation is the way to switch the resistance of RRAM devices.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

In Quest of Nonfilamentary Switching: A Synergistic Approach of Dual Nanostructure Engineering to Improve the Variability and Reliability of Resistive Random‐Access‐Memory Devices

Maikap

Banerjee

2020

Adv Elect Materials

View full text Add to dashboard Cite

This study demonstrates a synergistic approach of dual nanostructure engineering to improve the performance of AlOx‐based resistive random access memory (RRAM) devices. More precisely, a novel material stack engineering of RRAM with nonlinear area scalable low current switching and extensive improvement of reliability and variability issues is reported. Dual nanostructures (nanodome bottom electrode and nanocrystals in switching layer) in RRAM is an effective way to change the switching from filamentary to a nonfilamentary one, resulting in forming‐free highly stable device. Several methods such as transmission electron microscopy, energy dispersive spectroscopy, and atomic force microscopy have been employed to investigate the morphology of the RRAM devices. Enhanced electric field within the enclave of switching layer is controlling the intrinsic distribution of oxygen vacancy profile with extrinsic operation. As compare to filamentary RRAM, nonfilamentary devices can effectively control dc switching for >103 cycles, ac switching for >106 cycles without any severe fluctuation of resistance states for different levels. Area and temperature‐dependent semiconducting behavior along with time‐dependent high resistance states are certainly confirming the nonfilamentary switching process. In addition, the device yield improvement ≈98% using nonfilamentary switching, is making dual nanostructure devices a very promising candidate for high‐density memory application, is highlighted.

show abstract

“…Analog incremental current in S4 devices is also beneficial for neuromorphic type applications as compare to the abrupt current change in S2 devices. [ 2,16,25 ]…”

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

In Quest of Nonfilamentary Switching: A Synergistic Approach of Dual Nanostructure Engineering to Improve the Variability and Reliability of Resistive Random‐Access‐Memory Devices

Maikap

Banerjee

2020

Adv Elect Materials

View full text Add to dashboard Cite

show abstract

“…An ideal artificial synapse should present a linear and symmetric change in conductance under constant applied pulses for maximum accuracy, due to the nature of the algorithms used in neural networks (backpropagation algorithm) in pattern recognition. [54] However, the nonlinear response is not a serious issue thanks to learning algorithms which are dynamic and self-adjusting to errors in synapse programming. [55] The learning process in biological systems is established by the modification of synaptic weight or the plasticity mechanism, categorized as short-term and long-term memory.…”

Section: Synapse Behaviormentioning

confidence: 99%

Noble‐Metal‐Free Memristive Devices Based on IGZO for Neuromorphic Applications

Pereira

Deuermeier

Nogueira

et al. 2020

Adv Elect Materials

View full text Add to dashboard Cite

technology is everywhere. Currently, it is the Von Neumann's architecture that is applied to the electronic computing systems in which the different elements (memory, processor, and controller) are separated. [1] Nearly all the circuits within the memory and the processor are composed of complementary metal-oxidesemiconductor (CMOS) devices, which can be a problem once this technology cannot be further miniaturized without compromising its performance. [2] A practical example could be that in order to increase the operating frequency and the device density, which is necessary when downscaling, the power supply and the operation temperature would also increase, which would obviously degrade the system performance. Moreover, the Von Neumann system is not suited to solve real world problems where inputs and outputs are sometimes not specified [3] or to execute adaptive learning algorithms as it would be necessary in tasks, such as classification of unstructured data or pattern recognition. [4] To overcome the Von Neumann's bottleneck, the development of artificial intelligence technologies and new computer architecture designs are necessary. One of such novel approaches is neuromorphic computation, which operates with extremely low power consumption. It also maintains the massive parallelism found in the human brain [5] where neurons communicate information by electrical or chemical input signals passing through synapses. Synapses present a very important behavior called plasticity which consists in changing their strength (synaptic weight), either facilitating or inhibiting the connection between two neurons, through potentiation and depression, respectively. [6] One of the solutions to emulate biological synapses is the resistive switching (RS) device, or memristor. A memristor is basically a nonlinear two-terminal device whose conductance can be altered by external inputs and depends on the history of current that has flowed through the device. Application of both weight programming and weight processing signals are through its input terminals thus mimicking a synapse. In fact, the memristor can simulate the synapses' plasticity by continuously adapting its resistance into excitatory and inhibitory weights upon application of electrical Amorphous indium-gallium-zinc-oxide (a-IGZO) based memristive devices with molybdenum contacts as both top and bottom electrodes are presented aiming to be used in neuromorphic applications. Devices down to 4 µm 2 are fabricated using conventional photolithography processes, with an extraordinary yield of 100%. X-ray photoelectron spectroscopy and transmission electron microscopy performed on the developed structures confirm the presence of a thin intermixed oxide layer (4-5 nm) containing Mo 6+ oxidation state at the interface with the bottom contact. This results in Schottky diodelike characteristics at the pristine state with a rectification ratio of 3 orders of magnitude. The devices have electroforming-free and area-dependent analog resistive switching properties. Temperature...

show abstract

“…Besides the switching layer, the performance of non‐filamentary synapses also largely depends on electrode materials. For example, Moon et al obtained improved device uniformity, analog characteristics and retention in Mo/PCMO‐based synaptic device, where the improved retention can be explained by the high activation energy of the oxidation process and the high electronegativity of the Mo electrode. Similarly, other characteristics such as the current density and dynamic range can be improved in TiN/PCMO devices by varying sputtering conditions of the TiN electrode .…”

Section: Memristive Synapsesmentioning

confidence: 99%

Memristive Devices and Networks for Brain‐Inspired Computing

Zhang

et al. 2019

Physica Rapid Research Ltrs

View full text Add to dashboard Cite

As the era of big data approaches, conventional digital computers face increasing difficulties in performance and power efficiency due to their von Neumann architecture. As a result, there is recently a tremendous upsurge of investigations on brain-inspired neuromorphic hardware with high parallelism and improved efficiency. Memristors are considered as promising building blocks for the realization of artificial synapses and neurons and can therefore be utilized to construct hardware neural networks. Here, a review is provided on existing approaches for the implementation of artificial synapses and neurons based on memristive devices; and the respective advantages and disadvantages of these approaches are evaluated. This is followed by a discussion of hardware accelerators and neuromorphic computing systems that exploit the parallel, in-memory and analog characteristics of memristive crossbar arrays as well as the intrinsic dynamics of memristors. Finally, the outstanding challenges are addressed that have not yet been resolved in the present studies, and future advances are discussed that might be needed for building intelligent and energy efficient neuromorphic systems.

show abstract

RRAM-based synapse devices for neuromorphic systems

Abstract: We demonstrated a proton-based 3-terminal synapse device which shows symmetric conductance change characteristics. Using the optimized device, we successfully confirmed the improved classification accuracy of neural networks for on-chip training.

Cited by 192 publications

References 36 publications

In Quest of Nonfilamentary Switching: A Synergistic Approach of Dual Nanostructure Engineering to Improve the Variability and Reliability of Resistive Random‐Access‐Memory Devices

In Quest of Nonfilamentary Switching: A Synergistic Approach of Dual Nanostructure Engineering to Improve the Variability and Reliability of Resistive Random‐Access‐Memory Devices

Noble‐Metal‐Free Memristive Devices Based on IGZO for Neuromorphic Applications

Memristive Devices and Networks for Brain‐Inspired Computing

Contact Info

Product

Resources

About