内容寻址存储器
计算机科学
吞吐量
符号
算法
三元运算
并行计算
算术
数学
操作系统
人工智能
程序设计语言
人工神经网络
无线
作者
Ling Yang,Rongxuan Zhao,Yi Li,Hao Tong,Yingjie Yu,Xiangshui Miao
标识
DOI:10.1109/led.2022.3179736
摘要
Here, we proposed an in-memory search prototype based on phase change memory (PCM). First, using the PCM, a highly compact (8F 2 ) and low-energy (0.3 fJ/bit/search) nonvolatile ternary addressable memory (TCAM) is demonstrated and achieved a $6.6\times $ energy saving and $50\times $ cell area saving over the 16T SRAM-TCAM. Thanks to the non-volatility and massive parallelism of the PCM TCAM to computing Hamming distance, the frequent memory access is alleviated, enabling search operation in situ. The prototype shows a high throughput of 256 GB/s in data deduplication application, obtaining more than $250\times $ energy saving and $42\times $ improvement in throughput over CPU. Our study provides a low-area-overhead, low-energy, and high-throughput solution for the storage system to perform search in situ.
科研通智能强力驱动
Strongly Powered by AbleSci AI