2010 12th International Workshop on Cellular Nanoscale Networks and their Applications (CNNA 2010), Год журнала: 2023, Номер unknown, С. 1 - 4
Опубликована: Сен. 28, 2023
Язык: Английский
Scientific Reports, Год журнала: 2024, Номер 14(1)
Опубликована: Окт. 15, 2024
The adoption of transformer networks has experienced a notable surge in various AI applications. However, the increased computational complexity, stemming primarily from self-attention mechanism, parallels manner which convolution operations constrain capabilities and speed convolutional neural (CNNs). algorithm, specifically matrix-matrix multiplication (MatMul) operations, demands substantial amount memory thereby restricting overall performance transformer. This paper introduces an efficient hardware accelerator for network, leveraging memristor-based in-memory computing. design targets bottleneck associated with MatMul process, utilizing approximate analog computation highly parallel computations facilitated by memristor crossbar architecture. Remarkably, this approach resulted reduction approximately 10 times number multiply-accumulate (MAC) networks, while maintaining 95.47% accuracy MNIST dataset, as validated comprehensive circuit simulator employing NeuroSim 3.0. Simulation outcomes indicate area utilization 6895.7
Язык: Английский
Процитировано
0
Journal of Science Advanced Materials and Devices, Год журнала: 2024, Номер 9(4), С. 100805 - 100805
Опубликована: Окт. 24, 2024
Язык: Английский
Процитировано
0
ACS Applied Materials & Interfaces, Год журнала: 2024, Номер unknown
Опубликована: Дек. 4, 2024
The artificial photoelectric synaptic devices integrating optical signal response and data processing functions enable the simulation of human visual system. However, modulation behavior photoelectronic is susceptible to interference from sunlight background, significantly hindering their applications in modulation. To address this issue, a Ni/AlYN/ITO structure for sunlight-interference-free memristor was designed fabricated, application demonstrated. Solar-blind ultraviolet light with low background noise high security introduced as stimuli signal. Various biological functions, including short-term long-term plasticity, have been successfully simulated under electrical signals. Furthermore, it emulates system's perception memory exhibiting notable retention (over 300 s) capacity (up 5.6 times after 5 cycles). Notably, device's output current remains unaffected by interference. This device not only extends potential solar-blind region, but also facilitates integration communication brain-inspired computing. advancement expected information reception, transmission, resistant visible interference, thereby creating boosting secure communication.
Язык: Английский
Процитировано
0
FlexMat., Год журнала: 2024, Номер unknown
Опубликована: Дек. 16, 2024
Abstract Biological neural systems, composed of neurons and synaptic networks, exhibit exceptional capabilities in signal transmission, processing, integration. Inspired by the mechanisms these researchers have been dedicated to developing artificial systems based on flexible devices that effectively mimic functions biological synapses, providing hardware support for advancement intelligence. In recent years, ionic gels, known their high conductivity intuitive mimicry, utilized development ionic‐gel synapses (IGSs). They are considered ideal materials next wearable generation neuromorphic systems. This review introduces IGS summarizes progress IGS‐based Additionally, key challenges future prospects related IGSs outlined, potential suggestions provided.
Язык: Английский
Процитировано
0
2010 12th International Workshop on Cellular Nanoscale Networks and their Applications (CNNA 2010), Год журнала: 2023, Номер unknown, С. 1 - 4
Опубликована: Сен. 28, 2023
Язык: Английский
Процитировано
0