Proactive thermal management using memory based computing

摘要

Nanoscale devices provide the capability of gigascale integration in modern electronic systems. However, such systems suffer from high defect rates and large parametric variations. The surge of transistor count with the increased clock rate elevates the processor temperature which makes these systems even more unreliable and unstable. Dynamic Thermal Management (DTM) approaches considerably increase application's run-time in order to lower the peak temperature. Memory-based computing (MBC) is a promising approach to improve overall system reliability when few functional units are defective or unreliable under process-induced or thermal variations. In this paper, we present a novel DTM technique using proactive MBC to reduce the peak temperature of applications. We propose an efficient technique to proactively transfer the instructions with frequent operand pairs to memory. Experimental results demonstrate that the proposed proactive thermal management can significantly decrease the peak temperature to improve the system reliability with minor impact on performance.