Towards Energy-Efficient Computing Hardware Based on Memristive Nanodevices
Towards Energy-Efficient Computing Hardware Based on Memristive Nanodevices
复制标题
DOI：
10.1109/mnano.2023.3297106
复制
发表时间：
2023-10
影响因子：
1.6
通讯作者：
Y. Huang;Vignesh Ravichandran;Wuyu Zhao;Qiangfei Xia
中科院分区：
文献类型：
--
作者：
Y. Huang;Vignesh Ravichandran;Wuyu Zhao;Qiangfei Xia
计算硬件是影响我们日常生活的人工智能（AI）的关键驱动力之一。然而，尽管近几十年来取得了显著的进步，但为人工智能提供动力的计算硬件的能耗，特别是深度神经网络，仍然远远高于人类大脑的能耗。基于忆阻器等新兴纳米器件的硬件创新为节能计算系统提供了潜在的解决方案。这篇综述讨论了与开发基于忆阻纳米器件的节能计算硬件相关的挑战，并总结了忆阻器件、交叉阵列、系统和算法的最新进展，旨在从自下而上的方法解决这些问题。提出了进一步提高未来计算硬件能量效率的潜在研究方向。
Computing hardware is one of the crucial drivers of artificial intelligence (AI) that impacts our daily lives. However, despite the significant improvements made in recent decades, the energy consumption of computing hardware that powers AI, especially deep neural networks, remains considerably higher than that of human brains. Hardware innovations based on emerging nanodevices like memristors offer potential solutions to energy-efficient computing systems. This review discusses the challenges associated with developing energy-efficient computing hardware based on memristive nanodevices and summarizes recent progress in memristive devices, crossbar arrays, systems, and algorithms, aiming at addressing these issues from a bottom-up approach. Potential research directions are proposed to further improve future computing hardware's energy efficiency.