HR3AM: A Heat Resilient Design for RRAM-based Neuromorphic Computing
2019 IEEE/ACM International Symposium on Low Power Electronics and Design (ISLPED)(2019)
摘要
RRAM based accelerators have been widely adopted in many neuromorphic designs. However, RRAM cells are sensitive to temperature, which changes RRAM’s conductance. Such heat-induced interference can significantly decrease the computational accuracy because values are functions of RRAM conductance. In this paper, we propose HR
3
AM, a heat resilience design, which improves accuracy and optimizes the thermal distribution of RRAM based neural network accelerators. HR
3
AM consists of two key mechanisms: bitwidth downgrading and tile pairing. Bitwidth downgrading re-represents weights by shifting the conductance to improve the network inference accuracy. Tile pairing matches hot crossbar units with pre-defined idle units to mitigate high-temperature issues. We evaluated HR
3
AM on four real world neural network models. Results show that HR
3
AM improves classification accuracy by up to 41.8% compared with current state-of-the-art designs. For thermal optimization, HR
3
AM effectively decreases the maximum temperature by 6.2K and average temperature by 6K.
更多查看译文
关键词
hot crossbar units,real world neural network models,bitwidth downgrading,HR3AM,thermal optimization,high-temperature issues,pre-defined idle units,network inference accuracy,tile pairing,RRAM based neural network accelerators,heat resilience design,RRAM conductance,heat-induced interference,RRAM cells,neuromorphic designs,RRAM-based neuromorphic computing,heat resilient design
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要