The emergence of memristors with potential applications in data storage and artificial intelligence has attracted wide attentions. Memristors are assembled in crossbar arrays with data bits encoded by the resistance of individual cells. Despite the proposed high density and excellent scalability, the sneak-path current causing cross interference impedes their practical applications. Therefore, developing novel architectures to mitigate sneak-path current and improve efficiency, reliability, and stability may benefit next-generation storage-class memory (SCM). Moreover, conventional digital computers face the von-Neumann bottleneck and the slowdown of transistors’ scaling, imposing a big challenge to hardware artificial intelligence. Memristive crossbar features colocation of memory and processing units, as well as superior scalability, making it a promising candidate for hardware accelerating machine learning and neuromorphic computing. Herein, first, crossbar architecture is introduced. Then, for storage, the origin of sneak-path current is reviewed and techniques to mitigate this issue from the angle of materials and circuits are discussed. Computing wise, the applications of memristive crossbars in both machine learning and neuromorphic computing are surveyed, focusing on the structure of unit cells, the network topology, and the learning types. Finally, a perspective on future engineering and applications of memristive crossbars is discussed.

https://onlinelibrary.wiley.com/doi/pdf/10.1002/aisy.202100017

Memristive Crossbar Arrays for Storage and Computing Applications

Brain‐inspired associative memory is meaningful for pattern recognitions and image/speech processing. Here, a ferroelectric synaptic transistor network is proposed that is capable of associative learning and one‐step recalling of a whole set of data from only partial information. The competition between an external field and the internal depolarization field governs the ferroelectric creep of domain walls and offers each single ferroelectric synapse a full and subfemtojoule‐energy‐cost Hebbian synaptic plasticity, including short‐term memory (STM) to long‐term memory (LTM) transition, and remarkably both spike‐timing‐dependent plasticity (STDP) and spike‐rate‐dependent plasticity (SRDP). Assisted by the third terminal to control the ferroelectric domain dynamics, self‐adaptive coupling between neurons is realized by updating synaptic weight concurrently. Pavlov's dog experiment and multiassociative memories are demonstrated in this ferroelectric synaptic transistor network. Such ferroelectric synaptic transistor network is available for building multilayer neural networks and provides new avenues for associative‐memory information processing.

Ferroelectric Synaptic Transistor Network for Associative Memory

Abstract Nowadays the development of machine vision is oriented toward real-time applications such as autonomous driving. This demands a hardware solution with low latency, high energy efficiency, and good reliability. Here, we demonstrate a robust and self-powered in-sensor computing paradigm with a ferroelectric photosensor network (FE-PS-NET). The FE-PS-NET, constituted by ferroelectric photosensors (FE-PSs) with tunable photoresponsivities, is capable of simultaneously capturing and processing images. In each FE-PS, self-powered photovoltaic responses, modulated by remanent polarization of an epitaxial ferroelectric Pb(Zr 0.2 Ti 0.8 )O 3 layer, show not only multiple nonvolatile levels but also sign reversibility, enabling the representation of a signed weight in a single device and hence reducing the hardware overhead for network construction. With multiple FE-PSs wired together, the FE-PS-NET acts on its own as an artificial neural network. In situ multiply-accumulate operation between an input image and a stored photoresponsivity matrix is demonstrated in the FE-PS-NET. Moreover, the FE-PS-NET is faultlessly competent for real-time image processing functionalities, including binary classification between ‘X’ and ‘T’ patterns with 100% accuracy and edge detection for an arrow sign with an F-Measure of 1 (under 365 nm ultraviolet light). This study highlights the great potential of ferroelectric photovoltaics as the hardware basis of real-time machine vision. 

/pdf/ferroelectric-photosensor-network-an-advanced-hardware-3kdgeqm1.pdf

Ferroelectric photosensor network: an advanced hardware solution to real-time machine vision

DOI: 10.1002/admt.202100144 processing devices.[1–6] However, conventional image recognition systems using a flat image sensor array with a multilens optical system and the von-Neumann computing architecture for processing the acquired image data have several limitations such as high system-level complexity, bulky module size, large computing load, and low energy efficiency.[7] Therefore, advanced devices in both image acquisition and image data processing are required. As a result, bio-inspired imaging devices[8–10] (i.e., artificial vision) and neuromorphic image processing devices[11–13] (i.e., artificial synapse) have received considerable attention (Figure 1). Bio-inspired imaging devices (e.g., bioinspired camera) have been developed for image acquisition.[14] Conventional imaging devices require bulky and heavy optical systems to obtain high-quality visual information.[15] In contrast, natural eyes have a simple and small optical geometry and high-quality image acquisition capability.[16,17] Therefore, bio-inspired artificial vision has been developed by mimicking the unique structural and functional advantage of natural eyes[2,6] (Figure 1a). For example, the chambered eye, typically found in humans and aquatic animals, exhibits a wide field of view, low optical aberration, and facile accommodation with a simple optical system.[16,17] The compound eye has distinctive optical geometries, and such structures offer various useful visual features.[18,19] Neuromorphic computing devices that can efficiently process massive image data acquired from the imaging device have been developed for image data processing.[20–22] The conventional vonNeumann architecture, in which the central processing unit and memory unit are separated, is not suitable to efficiently process the massive unstructured image data.[23,24] Therefore, a novel computing device inspired by the human brain (i.e., electronic synapse) has been developed[25,26] (Figure 1b). For example, the memristor crossbar array can efficiently perform vector multiplications.[27] Such a neuromorphic device implements artificial neural networks (ANN) in the hardware and enables efficient parallel processing of image data with low energy consumption.[25] In a previous study, a device that integrates the synaptic device and photodetector in one unit has been reported.[26] Despite recent progress in the hardware of the neuromorphic image data processing devices, such devices still require Remarkable technological developments for efficient image recognition (i.e., image acquisition and image data processing) have been reported in the past decade. Such advances in imaging and image processing technologies have driven significant progress in mobile electronics and machine vision applications. In particular, for image acquisition devices, two types of natural eyes (i.e., chambered and compound eyes) have inspired the development of novel multifunctional imaging devices with unique optical geometries. For image data processing devices, novel computing devices based on memristor crossbar arrays, such as electronic synapses, have been developed. More recently, the integration of imaging and image processing devices in a single unit further enhances the system-level efficiency. Herein, such recent advances in the bio-inspired artificial vision and neuromorphic image processing devices, aimed at providing efficient image recognition, are reviewed. First, various imaging devices inspired by the structural and functional features of natural eyes are introduced. Second, artificial synapses and their operation principles are thoroughly discussed. Third, the neuromorphic vision sensor that integrates the imaging and image processing devices is reviewed. Finally, a brief summary and future outlook are presented.

Bio-Inspired Artificial Vision and Neuromorphic Image Processing Devices

Developing a single-phase self-rectifying memristor with the continuously tunable feature is structurally desirable and functionally adaptive to dynamic environmental stimuli variations, which is the pursuit of further smart memristors and neuromorphic computing. Herein, we report a van der Waals ferroelectric CuInP2S6 as a single memristor with superior continuous modulation of current and self-rectifying to different bias stimuli (sweeping speed, direction, amplitude, etc.) and external mechanical load. The synergetic contribution of controllable Cu+ ions migration and interfacial Schottky barrier is proposed to dynamically control the current flow and device performance. These outstanding sensitive features make this material possible for being superior candidate for future smart memristors with bidirectional operation mode and strong recognition to input faults and variations. 

/pdf/manipulation-of-current-rectification-in-van-der-waals-20131o85.pdf

Manipulation of current rectification in van der Waals ferroionic CuInP2S6

Since the emergence of memristors (or memristive devices), how to integrate them into arrays has been widely investigated. After years of research, memristor crossbar arrays have been proposed and realized with potential applications in nonvolatile memory, logic and neuromorphic computing systems. Despite the promising prospects of memristor crossbar arrays, one of the main obstacles for their development is the so-called sneak-path current causing cross-talk interference between adjacent memory cells and thus may result in misinterpretation which greatly influences the operation of memristor crossbar arrays. Solving the sneak-path current issue, the power consumption of the array will immensely decrease, and the reliability and stability will simultaneously increase. In order to suppress the sneak-path current, various solutions have been provided. So far, some reviews have considered some of these solutions and established a sophisticated classification, including 1D1M, 1T1M, 1S1M (D: diode, M: memristor, T: transistor, S: selector), self-selective and self-rectifying memristors. Recently, a mass of studies have been additionally reported. This review thus attempts to provide a survey on these new findings, by highlighting the latest research progress realized for relieving the sneak-path issue. Here, we first present the concept of the sneak-path current issue and solutions proposed to solve it. Consequently, we select some typical and promising devices, and present their structures and properties in detail. Then, the latest research activities focusing on single-device structures are introduced taking into account the mechanisms underlying these devices. Finally, we summarize the properties and perspectives of these solutions.

/pdf/research-progress-on-solutions-to-the-sneak-path-issue-in-4khx7eisto.pdf

Lingyun Shi

Papers

Research progress on solutions to the sneak path issue in memristor crossbar arrays