A Novel Scalable Energy-Efficient Synaptic Device: Crossbar Ferroelectric Semiconductor Junction

Si, Mengwei; Ye, P. D.; Luo, Yuan-Chun; Chung, Wan-Hsuan; Bae, Hagyoul; Zheng, Dongqi; Li, J.; Qin, Jiaqian; Qiu, Gang; Yu, Shimeng

doi:10.1109/iedm19573.2019.8993622

Cited by 25 publications

(47 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To mimic the functions of biological neurons and synapses, the synapse-like electronic devices based on various 2D memory technologies have been explored. [52,57,[189][190][191][192][193][194][195][196][197][198][199] The volatile and nonvolatile retention characteristics of a memory device are natural emulation of STP and LTP of a biological synapse, respectively. Moreover, synaptic connection weight can be represented by memory conductance.…”

Section: Synaptic Device Implementationsmentioning

confidence: 99%

“…For example, Si et al reported the fabrication of energy-efficient ferroelectric synaptic devices based on α-In 2 Se 3 flakes. [198] Vertical voltages can adjust the height and width of Schottky barrier at the metal/semiconductor interface by controlling the out-of-plane ferroelectric polarization. Therefore, the potentiation and depression of the synaptic devices can be realized via applying successive pulses.…”

Section: Spike-rate-dependent Plasticity (Srdp)mentioning

confidence: 99%

See 1 more Smart Citation

Emerging 2D Memory Devices for In‐Memory Computing

et al. 2021

View full text Add to dashboard Cite

cube (HMC), [5] high bandwidth memory (HBM), [6,7] and 3D monolithic integration, [8] have been successively developed and achieved some success, the challenge of latency and energy consumption caused by massive data transmission still remains. Therefore, it is urgently necessary to enhance the information interaction capability by improving the architectural relationship between processing and memory units.The emergence of non-von-Neumann architecture solves the problem of the separation of processing and memory units, and alleviates the impact of bus bandwidth on computing efficiency. As a non-Von Neumann architecture, in-memory computing has been considered to be one of the future mainstream trends of hardware implementation for AI algorithms, [9] such as machine learning and deep learning. In-memory computing, where calculations are carried out in situ within each memory unit, [10] has massive parallelism and distributed computing characteristic through integrating millions of memory devices in a crossbar array (Figure 1), analogous to the neurobiological system. [11] Thus, it can radically subvert the von Neumann architecture and achieve the fusion of data processing and storage, thereby totally eliminating the latency and energy loss associated with data access. For enabling this computational architecture, new material systems and highperformance memory devices are highly pursued.2D layered materials refer to the material family that held together by strong in-plane chemical bonds and relatively weak out-of-plane van der Waals interactions, and have attracted worldwide attention due to their unique structure and physical properties. [12][13][14][15][16][17][18][19][20][21] On the one hand, the atomically thin thickness of 2D layered materials provide a significant advantage in achieving high-density integration and low-power operation of high-performance devices. [22][23][24][25][26][27][28] On the other hand, their dangling-bond-free surface and planar structure make them not only compatible with traditional wafer technology, but also can be stacked on top of each other unrestricted by lattice mismatch. Thus, varieties of available 2D layered materials with different electrical properties, such as graphene, transition metal dichalcogenides (TMDs, including MoS 2 , WSe 2 , MoTe 2 , etc.), black phosphorus (BP) and hexagonal boron nitride (h-BN), could be arbitrarily assembled to create a wide range of artificial van der Waals heterostructures (vdWHs) with wholly new functionalities that unavailable in the individual material. [29][30][31][32][33] For example, stable data storage can be realized with the aid of potential barrier formed in vdWHs. [34][35][36][37][38][39] Additionally, It is predicted that the conventional von Neumann computing architecture cannot meet the demands of future data-intensive computing applications due to the bottleneck between the processing and memory units. To try to solve this problem, in-memory computing technology, where calculations are carried out in situ within each nonv...

show abstract

Section: Synaptic Device Implementationsmentioning

confidence: 99%

Section: Spike-rate-dependent Plasticity (Srdp)mentioning

confidence: 99%

Emerging 2D Memory Devices for In‐Memory Computing

et al. 2021

View full text Add to dashboard Cite

show abstract

“…They could overcome the challenges faced by traditional ferroelectrics and have good applications in ultrahigh density and ultrafast speed memory, [ 30–33 ] optoelectronic multifunctional memories, [ 34,35 ] and ferroelectric semiconductor field‐effect transistors. [ 3,36,37 ] These materials usually have polar point group structure, such as 3m α‐In 2 Se 3 and m CuInP 2 S 6 . Yet, to introduce polar structure into otherwise nonpolar 2D materials to produce ferroelectricity, which enables more functionality and more flexible controllability, remains a challenge.…”

Section: Introductionmentioning

confidence: 99%

Discovery of Robust Ferroelectricity in 2D Defective Semiconductor α‑Ga₂Se₃

Xue

Jiang

Wang

et al. 2021

Small

View full text Add to dashboard Cite

2D ferroelectrics with robust polar order in the atomic‐scale thickness at room temperature are needed to miniaturize ferroelectric devices and tackle challenges imposed by traditional ferroelectrics. These materials usually have polar point group structure regarding as a prerequisite of ferroelectricity. Yet, to introduce polar structure into otherwise nonpolar 2D materials for producing ferroelectricity remains a challenge. Here, by combining first‐principles calculations and experimental studies, it is reported that the native Ga vacancy‐defects located in the asymmetrical sites in cubic defective semiconductor α‑Ga2Se3 can induce polar structure. Meanwhile, the induced polarization can be switched in a moderate energy barrier. The switched polarization is observed in 2D α‑Ga2Se3 nanoflakes of ≈4 nm with a high switching temperature up to 450 K. Such polarization switching could arise from the displacement of Ga vacancy between neighboring asymmetrical sites by applying an electric field. This work removes the point group limit for ferroelectricity, expanding the range of 2D ferroelectrics into the native defective semiconductors.

show abstract

“…In addition, as α-In 2 Se 3 is not an oxide, the issues related to oxygen vacancies in oxide Fe-insulator are expected to be non-existent in this FeS material. Recently, similar to FTJ, a metal-FeSmetal junction device (called FeSMJ) has been demonstrated to exhibit polarization-dependent resistance states 17 .…”

mentioning

confidence: 99%

“…Unlike FTJ, FeSMJ can provide significant current density even with a high FeS thickness and it does not require asymmetric metal electrodes for NVM functionalities 17 . To understand such unique working principle of FeSMJ devices and to enable their device-level optimization, a detailed analysis of the material properties α-In 2 Se 3 as well as the device characteristics of FeSMJ, is needed.…”

mentioning

confidence: 99%

α -In2Se3 based ferroelectric-semiconductor metal junction for non-volatile memories

Saha

et al. 2020

Applied Physics Letters

View full text Add to dashboard Cite

In this work, we theoretically and experimentally investigate the working principle and non-volatile memory (NVM) functionality of 2D α-In 2 Se 3 based ferroelectric-semiconductor-metal-junction (FeSMJ). First, we analyze the semiconducting and ferroelectric properties of α-In 2 Se 3 van-der-Waals (vdW) stack via experimental characterization and first-principle simulations. Then, we develop a FeSMJ device simulation framework by selfconsistently solving Landau-Ginzburg-Devonshire (LGD) equation, Poisson's equation, and charge-transport equations. Based on the extracted FeS parameters, our simulation results show good agreement with the experimental characteristics of our fabricated α-In 2 Se 3 based FeSMJ. Our analysis suggests that the vdW gap between the metal and FeS plays a key role to provide FeS polarization-dependent modulation of Schottky barrier heights. Further, we show that the thickness scaling of FeS leads to a reduction in read/write voltage and an increase in distinguishability. Array-level analysis of FeSMJ NVM suggests a 5.47x increase in sense margin, 18.18x reduction in area and lower read-write power with respect to Fe insulator tunnel junction (FTJ). Ferroelectric (Fe) materials have gained an immense research interest for their applications in electronic 1-3 and optoelectronic 4 devices due to their electrically switchable spontaneous polarization and hysteretic characteristics. Fe materials with high bandgap, called Fe-insulators, have been extensively investigated for versatile non-volatile memory (NVM) devices, such as Fe-random-access memory (Fe-RAM) 5 , Fe-field-effecttransistor (Fe-FET) 6-7 , Fe-tunnel-Junction (FTJ) 8-10 , etc. Unlike Fe-RAM and Fe-FET, where the Fe layer acts as a capacitive element, the FTJ functionality depends on the tunneling current through the Fe layer. In the FTJ, the Fe layer is sandwiched by two different metal electrodes. Due to the different properties (e.g. the screening length) of the metal electrodes, the tunneling barrier height at the metal-Fe interface of FTJ depends on the polarization direction. Thus, FTJ can exhibit polarization-dependent tunneling-resistance that facilitates the sensing of its polarization, leading to the design of a two-terminal NVM element 8 . However, as the dominant transport mechanism of FTJ is direct tunneling, to obtain a desired current density for sufficient operational speed, the Feinsulator thickness needs to be significantly low (<3nm for HZO 9 ). Unfortunately, with thickness scaling, the polarization of the Fe-insulator decreases 9 , which reduces the ratio of the tunneling-resistance 9 . Therefore, the distinguishability of the FTJ memory states decreases with the thickness scaling. In addition, most of the Feinsulators (BTO, BFO, doped-HfO 2 ) comprise of oxygen atoms and the dynamic change in oxygen vacancies can play a major role in its ferroelectric characteristics 11 . Therefore, a decrease in ferroelectricity with scaling and issues related to oxygen vacancies lead to significant challenges in the design and...

show abstract

A Novel Scalable Energy-Efficient Synaptic Device: Crossbar Ferroelectric Semiconductor Junction

Cited by 25 publications

References 13 publications

Emerging 2D Memory Devices for In‐Memory Computing

Emerging 2D Memory Devices for In‐Memory Computing

Discovery of Robust Ferroelectricity in 2D Defective Semiconductor α‑Ga₂Se₃

α -In2Se3 based ferroelectric-semiconductor metal junction for non-volatile memories

Contact Info

Product

Resources

About

A Novel Scalable Energy-Efficient Synaptic Device: Crossbar Ferroelectric Semiconductor Junction

Cited by 25 publications

References 13 publications

Emerging 2D Memory Devices for In‐Memory Computing

Emerging 2D Memory Devices for In‐Memory Computing

Discovery of Robust Ferroelectricity in 2D Defective Semiconductor α‑Ga2Se3

α -In2Se3 based ferroelectric-semiconductor metal junction for non-volatile memories

Contact Info

Product

Resources

About

Discovery of Robust Ferroelectricity in 2D Defective Semiconductor α‑Ga₂Se₃