Computational Restructuring: Rethinking Image Processing using Memristor Crossbar Arrays
Computational Restructuring: Rethinking Image Processing using Memristor Crossbar Arrays
复制标题
DOI：
10.23919/date48585.2020.9116255
复制
发表时间：
2020-03
期刊：
影响因子：
--
通讯作者：
Baogang Zhang;Necati Uysal;Rickard Ewetz
中科院分区：
文献类型：
--
作者：
Baogang Zhang;Necati Uysal;Rickard Ewetz
图像处理是对物联网（IoT）中数十亿个传感器设备执行的核心操作。新兴的Memristor横梁阵列（MCAS）有望用极小的能量延迟产品执行矩阵矢量乘法（MVM），这是二维离散余弦变换（2D DCT）中的主导计算。较早的研究已将数字实施直接映射到基于MCA的硬件。缺点是该系列计算容易受到错误的影响。此外，实现需要使用大图像块尺寸，这是已知可以降低图像质量的。在本文中，我们建议将2D DCT重组为等效的单线性转换（或MVM操作）。重建消除了系列计算，并将处理后的块大小从NXN降低到$ \ sqrt N {\ MathBf {x}}} \ sqrt n $，因此，对错误的稳健性和图像质量的稳健性都得到改善。此外，使用频率频谱优化，使用中间数据的存储在消除中间数据的同时减小了潜伏期，功率和面积，并消除了中间数据的存储，最多可将功率和区域进一步降低62％和74％。
Image processing is a core operation performed on billions of sensor-devices in the Internet of Things (IoT). Emerging memristor crossbar arrays (MCAs) promise to perform matrix-vector multiplication (MVM) with extremely small energy-delay product, which is the dominating computation within the two-dimensional Discrete Cosine Transform (2D DCT). Earlier studies have directly mapped the digital implementation to MCA based hardware. The drawback is that the series computation is vulnerable to errors. Moreover, the implementation requires the use of large image block sizes, which is known to degrade the image quality. In this paper, we propose to restructure the 2D DCT into an equivalent single linear transformation (or MVM operation). The reconstruction eliminates the series computation and reduces the processed block sizes from NxN to $\sqrt N {\mathbf{x}}\sqrt N $ Consequently, both the robustness to errors and the image quality is improved. Moreover, the latency, power, and area is reduced with 2X while eliminating the storage of intermediate data, and the power and area can be further reduced with up to 62% and 74% using frequency spectrum optimization.