Article
Computer Science, Information Systems
Xingqi Zou, Sheng Xu, Xiaoming Chen, Liang Yan, Yinhe Han
Summary: Processing-in-memory (PIM) is proposed as a promising solution to break the von Neumann bottleneck by minimizing data movement between memory hierarchies. This study focuses on prior art of architecture level DRAM PIM technologies and their implementation, discussing the key challenges, mainstream solutions, relative limitations of PIM simulation, and conventional PIM simulators. Research directions and perspectives are proposed for future development in PIM.
SCIENCE CHINA-INFORMATION SCIENCES
(2021)
Article
Nanoscience & Nanotechnology
Jia-Qin Yang, Ye Zhou, Su-Ting Han
Summary: The development of artificial intelligence and big data analytics is revolutionizing data processing and storage methods, requiring new computing systems to address speed and energy consumption issues. Advances in various data storage technologies have made functional computing possible, while an overview of current memory systems and their development history provides required figure of merits for applications like in-memory computing and identifies critical issues for future development.
ADVANCED ELECTRONIC MATERIALS
(2021)
Article
Computer Science, Theory & Methods
Leonid Yavits, Roman Kaplan, Ran Ginosar
Summary: GIRAF is a general framework that combines storage and parallel associative processing using resistive content addressable memory. It improves performance by conducting computations within the storage arrays and addresses bandwidth limitations.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
(2022)
Review
Computer Science, Information Systems
Caidie Cheng, Pek Jun Tiw, Yimao Cai, Xiaoqin Yan, Yuchao Yang, Ru Huang
Summary: In-memory computing represents a radical shift in computer architecture, addressing fundamental limitations in latency and energy consumption posed by the von Neumann bottleneck and memory wall. This article reviews emerging nonvolatile memory devices and discusses the optimizations required at the device and array levels to better support in-memory computing. Additionally, recent progress in applying in-memory computing in artificial neural networks, spiking neural networks, digital logic in memory, and hardware security is discussed, along with remaining challenges in the field and potential pathways to address them.
SCIENCE CHINA-INFORMATION SCIENCES
(2021)
Article
Chemistry, Multidisciplinary
Yujie Song, Xingsheng Wang, Qiwen Wu, Fan Yang, Chengxu Wang, Meiqing Wang, Xiangshui Miao
Summary: This study proposes a simple and cost-effective logic scheme that allows arbitrary Boolean logic calculation using only two memristors and one resistor. The proposed scheme enables the implementation of 1-bit and N-bit adder circuits and addresses critical issues, supporting the development of a high-functional reconfigurable digital in-memory calculation system.
Article
Physics, Applied
H. S. Alagoz, M. Egilmez, J. Jung, K. H. Chow
Summary: We studied the electrical and thermal conduction properties of Pt/NiOx/Pt based unipolar ReRAM devices in low- (ON) and high-resistance (OFF) states during cooling and warming cycles between 300 and 180 K. The activation energy of conduction electron-trap decreased upon warming. Thermal cycling didn't significantly affect the average resistance-temperature coefficient of the Pt diffused conductive filaments, but it increased the ON-state resistance fluctuations at high temperatures, indicating the influence of ambient temperature on the sizes of formed filaments. The mechanism behind these thermally activated changes is discussed.
APPLIED PHYSICS LETTERS
(2023)
Article
Computer Science, Information Systems
Soonbum Song, Youngmin Kim
Summary: This study introduces an 8(+)T SRAM IMC circuit and proposes an IMC full adder and approximate adder based on this circuit. The circuit can perform read and compute operations simultaneously, contributing to reduced energy consumption and improved overall performance.
Article
Engineering, Electrical & Electronic
Chandan Kumar Jha, Sallar Ahmadi-Pour, Rolf Drechsler
Summary: In this paper, a library of 16 by 8 approximate restoring array dividers suitable for memristor-based in-memory computing (IMC) is proposed. The library can be used as benchmark circuits. The proposed designs are compared with existing CMOS-based approximate divider designs when mapped to IMCs, and a case study on two image processing applications is conducted.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2023)
Article
Engineering, Electrical & Electronic
Orian Leitersdorf, Ronny Ronen, Shahar Kvatinsky
Summary: Processing-in-memory (PIM) eliminates computation/memory data transfer by using devices that support both storage and logic. The state-of-the-art algorithm for stateful single-row multiplication, RIME, reduces latency by 5.1x using memristive partitions. This brief presents novel partition-based computation techniques, an in-memory multiplication algorithm based on CSAS technique, and a stateful full-adder that improves the state-of-the-art design. These contributions constitute MultPIM, a multiplier that reduces time complexity from quadratic to linear-log. For 32-bit numbers, MultPIM reduces latency by an additional 4.2x over RIME, while slightly reducing area overhead. Furthermore, MultPIM is optimized for full-precision matrix-vector multiplication and improves latency by 25.5x over FloatPIM matrix-vector multiplication.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2022)
Article
Computer Science, Artificial Intelligence
Yongbin Yu, Jiehong Mo, Quanxin Deng, Chen Zhou, Biao Li, Xiangxiang Wang, Nijing Yang, Qian Tang, Xiao Feng
Summary: This article proposes a matrix-friendly genetic algorithm (MGA), which achieves population evolution through matrix operations. Compared to a baseline genetic algorithm (GA), MGA shows better and faster convergence. Utilizing the parallelism of matrix operations, MGA runs 2.5 times faster when using the NumPy library. Additionally, the article introduces the deployment of MGA using memristor circuits, enabling parallelization and in-memory computing. Experimental results demonstrate that feature selection based on MGA can achieve higher accuracy in logistic regression.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Engineering, Electrical & Electronic
Mohsen Riahi Alam, M. Hassan Najafi, Nima TaheriNejad, Mohsen Imani, Raju Gottumukkala
Summary: Stochastic Computing (SC) is an alternative computing paradigm that offers high robustness to noise and outstanding efficiency compared to traditional binary computing. However, the conversion between binary and stochastic representation comes at a significant cost with current CMOS technology. In-Memory Computation (IMC) is introduced to speed up Big Data applications and provide massive parallelism. This work explores the use of IMC, particularly with memristors, to design fast and energy-efficient SC systems.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2022)
Article
Computer Science, Hardware & Architecture
Souvik Kundu, Priyanka B. Ganganaik, Jeffry Louis, Hemanth Chalamalasetty, B. V. V. S. N. Prabhakar Rao
Summary: This article proposes an innovative memristor crossbar-based architecture, CoCoPIM, for computing correlation parameters in-memory, which significantly improves the energy efficiency and speedup compared to traditional von Neumann machine. The performance of CoCoPIM was evaluated in various applications, showing remarkable energy efficiency and speedup.
IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS
(2022)
Article
Chemistry, Multidisciplinary
Mohit Kumar, Yeong Hwan Ahn, Shahid Iqbal, Unjeong Kim, Hyungtak Seo
Summary: Emerging nonvolatile resistive switching, or memristors, rely on the composition change of active materials rather than storing charge. By stabilizing the conduction channel through manipulation of oxygen defects using electron-beam irradiation, high-performance memristor devices with adjustable switching ratios are fabricated, showcasing broad modulation of neural activities.
Article
Engineering, Electrical & Electronic
Xiaoyue Ji, Zhekang Dong, Chun Sing Lai, Donglian Qi
Summary: The study introduces a brain-inspired multimodal signal processing system that integrates signal sensory, storage, and computation through organic memristor arrays. The system consists of four key components: multimodal signal sensory module, high-density cross-point memristive synapse array, general learning module, and peripheral circuit module. The system is able to capture massive amounts of data every second and perform in situ processing of multimodal signals, potentially advancing the deep integration of nano materials into neuromorphic computing systems and energy-efficient integrated circuits.
IEEE COMMUNICATIONS MAGAZINE
(2022)
Article
Multidisciplinary Sciences
Markus Hellenbrand, Babak Bakhit, Hongyi Dou, Ming Xiao, Megan O. Hill, Zhuotong Sun, Adnan Mehonic, Aiping Chen, Quanxi Jia, Haiyan Wang, Judith L. MacManus-Driscoll
Summary: A design concept of phase-separated amorphous nanocomposite thin films is presented that enables interfacial resistive switching in hafnium oxide-based devices. By incorporating an average of 7% Ba into hafnium oxide, films consisting of amorphous HfOx host matrix interspersed with Ba-rich nanocolumns are formed. The resistive switching is restricted to an interfacial Schottky-like energy barrier, which can be tuned by ionic migration under an applied electric field. The resulting devices exhibit stable reproducibility and switching endurance, allowing for multiple intermediate resistance states and synaptic spike-timing-dependent plasticity.