Optimizing Method For Chip Multiprocessor In The Dark Silicon Era

CMPs(Chip MultiProcessors)have been widely concerned and used.With the development of technology node in semiconductor manufacturing,the on-chip power density and heat generation of the chips are rapidly increasing.Therefore,CMPs can’t active all processors with maximum frequency at the same time.It’s the cause of dark silicon.This paper concerns on optimizing computational performance and security of CMP in dark silicon era.The main contents are as follow:(1)The key concepts and techniques are introducedAt first,The concepts of TDP(thermal design power)and hardware security are introduced.Then,The performance optizimizing techniques such as DVFS(Dynamic Voltage & Frequency Scaling),computational sprinting and task mapping are introduced.Furthermore,The security optimizing techniques such as trojan detecting,task scheduling and noise generating are also introduced.(2)The optimizing method based on computational sprinting is proposedA new optimizing method for computational performance is proposed in this paper.At first,constraints of the optimizing pattern(best voltage/frequency)is analyzed.Then,a mathematical model is built to prove the rationality of the optimizing pattern.Finally,a searching algorithm is built.Giving TDP and standard voltage/frequency levels,the algorithm can automatically find the best pattern for certain application.The simulation results revel that the optimizing pattern propsed in this paper,outperforms all other sprinting pattern either on computational performance and energy efficiency.(3)The optimizing method of security performance based on task mapping is establishedA new security-oriented task mapping strategy for thermal side channel leakage is proposed in this paper.At first,the thermal distribution of CMP is analyzed.Then,a cost function based on processors’ usage,temperature and location is built based on above analyzing.At last,a security-oriented mapping algorithm based on above cost function is established.The algorithm can dynamically assign application threads to cores with minimal cost principle.The simulation results revel that,our security-oriented algorithm can effectively reduce the thermal side channel leakage comparing with existing task mapping algorithms.At the end of this paper,above works are summarized,disadvantages are analyzed and future work is prospected respectively.