计算机科学
功率消耗
绘图
炸薯条
GSM演进的增强数据速率
检测点注意事项
嵌入式系统
计算机硬件
能源消耗
功率(物理)
人工智能
实时计算
工程类
计算机图形学(图像)
电气工程
电信
医学
免疫学
物理
量子力学
作者
Xu Zheng,Lizhou Wu,Yixuan Liu,Qi-Qiao Wu,Yuanlu Xie,Yi Li,Jinru Lai,Wenxuan Sun,Danian Dong,Jie Yu,Xiaoxin Xu,Wenchang Zhang,Ming Liu
标识
DOI:10.1109/iedm45741.2023.10413719
摘要
The Point-of-Care Testing (POCT) system based on image analysis suffers high power consumption, detection delay and lack of privacy security. In this work, we proposed an efficient POCT system based on memristor chip, which can support edge computing and realize real-time recognition of medical graphics. To solve the problem of the decrease of long-term inference accuracy caused by chip degradation, we designed the error correction and in-situ recovery module, which increased the recognition accuracy to 90% after 250 days. Benefit by the high speed and low power consumption of the memristor, the efficiency of POCT system is up to 1.547TOPS/W and the recognition rate is reach to 1142 frames/s.
科研通智能强力驱动
Strongly Powered by AbleSci AI