Design and Optimization of Resistive RAM-based Storage and Computing Systems
Year: 2019
Type: dissertation
Abstract: : The Resistive Random Access Memory (ReRAM) is an emerging non-volatile memory\n\ntechnology because of its attractive attributes, including excellent scalability (< 10 nm), low\n\nprogramming voltage (< 3 V), fast switching speed (< 10 ns), high OFF/ON ratio (> 10),\n\ngood endurance (up to 1012 cycles) and great compatibility with silicon CMOS technology [1].\n\nHowever, ReRAM suffers from larger write latency, energy and reliability issue compared to\n\nDynamic Random Access Memory (DRAM). To improve the energy-efficiency, latency efficiency and reliability of ReRAM storage systems, a low cost cross-layer approach that spans device, circuit, architecture and system levels is proposed.\n\nFor 1T1R 2D ReRAM system, the effect of both retention and endurance errors on\n\nReRAM reliability is considered. Proposed approach is to design circuit-level and architecture-level techniques to reduce raw Bit Error Rate significantly and then employ low cost Error Control Coding to achieve the desired lifetime.\n\nFor 1S1R 2D ReRAM system, a cross-point array with “multi-bit per access” per subarray\n\nis designed for high energy-efficiency and good reliability. The errors due to cell-level as well\n\nas array-level variations are analyzed and a low cost scheme to maintain reliability and latency\n\nwith low energy consumption is proposed.\n\nFor 1S1R 3D ReRAM system, access schemes which activate multiple subarrays with\n\nmultiple layers in a subarray are used to achieve high energy efficiency through activating fewer\n\nsubarray, and good reliability is achieved through innovative data organization.\n\nFinally, a novel ReRAM-based accelerator design is proposed to support multiple\n\nConvolutional Neural Networks (CNN) topologies including VGGNet, AlexNet and ResNet.\n\nThe multi-tiled architecture consists of 9 processing elements per tile, where each tile \n\nimplements the dot product operation using ReRAM as computation unit. The processing\n\nelements operate in a systolic fashion, thereby maximizing input feature map reuse and\n\nminimizing interconnection cost. The system-level evaluation on several network benchmarks\n\nshow that the proposed architecture can improve computation efficiency and energy efficiency\n\ncompared to a state-of-the-art ReRAM-based accelerator.
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7046168501
