This paper proposes a novel spike generator for processing in memory (PIM) technology. Most of the electronics today utilize a von Neumann architecture. The von Neumann architecture suffers from the separation of memo...
详细信息
With advances in neural network technology, Processing-In-Memory (PIM) has emerged as a solution to per-formance bottlenecks between processors and memory. Among various PIM design techniques, integrating processing u...
详细信息
This paper analyzes the impact of Direct Cache Access (DCA) within a multicore Non-Uniform Memory Access (NUMA) framework on network packet processing efficiency in Linux systems. By strategically placing network flow...
详细信息
This paper presents a novel design technique for reducing ripple in various sensor read-out integrated circuits (IC) by calibrating the amplifier offset through current control in bias resistors. Traditional methods f...
详细信息
作者：
Butola, RajatLi, YimingKola, Sekhar ReddyNational Yang Ming Chiao Tung University
Parallel and Scientific Computing Laboratory Electrical Engineering and Computer Science International Graduate Program Hsinchu300093 Taiwan Institute of Pioneer Semiconductor Innovation
The Institute of Artificial Intelligence Innovation National Yang Ming Chiao Tung University Parallel and Scientific Computing Laboratory Electrical Engineering and Computer Science International Graduate Program The Institute of Communications Engineering the Institute of Biomedical Engineering Department of Electronics and Electrical Engineering Hsinchu300093 Taiwan
In this work, a dynamic weighting-artificial neural network (DW-ANN) methodology is presented for quick and automated compact model (CM) generation. It takes advantage of both TCAD simulations for high accuracy and SP...
详细信息
3D U-Net medical image segmentation is essential for precise diagnosis and treatment planning but comes with massive bandwidth requirements and considerable computational costs. Eight-bit fixed-point quantization is c...
详细信息
This study investigates the potential of NucleoBases (NBs) materials, with a specific focus on Thymine, as HTL for QDLEDs. NBs materials offer cost-effectiveness compared to traditional organic optoelectronic material...
详细信息
This study integrates an inductor into OLED structures to overcome challenges in low gray level expression. At the resonance frequency of 19 kHz, a linear correlation between voltage and emission intensity emerges, re...
详细信息
In inverted quantum-dot light-emitting-diodes (QD-LED), it is important to use orthogonal solvent for hole-transport layer(HTL). In addition, modulating work function of each layer is good way to enhance carrier injec...
详细信息
With advances in neural network technology, Processing-In-Memory (PIM) has emerged as a solution to per-formance bottlenecks between processors and memory. Among various PIM design techniques, integrating processing u...
详细信息
ISBN:
(数字)9798331510756
ISBN:
(纸本)9798331510763
With advances in neural network technology, Processing-In-Memory (PIM) has emerged as a solution to per-formance bottlenecks between processors and memory. Among various PIM design techniques, integrating processing units within memory banks has demonstrated high performance in accelerating neural network models. However, prior architectures based on this approach have required sacrificing a large portion of the cell array to accommodate complex floating-point pro-cessing units. In this paper, we propose a new architecture that incorporates a unique quantization technique called microscaling. This technique efficiently converts high-precision data to integer types with minimal accuracy loss during model inference. By adopting microscaling, we reduce the processing unit overhead by replacing floating-point units with integer-based ones. As a result, our approach achieves a 50 % reduction in area while maintaining equivalent performance and reduces energy consumption to approximately 70 % of that in prior architectures.
暂无评论