Recurrent Neural Network Based Dynamic Thermal Management For Multi-Core Systems

Leakage power is becoming significant in new generation IC chips.As leakage power is nonlinearly related to temperature,it is challenging to manage the thermal behavior of today's multi-core systems,since thermal management becomes a nonlinear control problem.In this paper,research topic is the recurrent neural network based dynamic thermal management(DTM)for multi-core systems.It is proposed to use the recurrent neural network(RNN)method to build leakage-aware thermal model for multi-core systems,and then use recurrent neural network based model prediction control(MPC)method to management temperature of chips.Firstly,the paper analyzes the interdependence problem in the traditional thermal model based dynamic thermal management method when considering the nonlinear relationship between leakage power and temperature,where we need to know the leakage power to calculate the temperature and also need the temperature to calculate the leakage power.Then,in order to find a practical leakage-aware thermal model for the dynamic thermal management,we innovatively designed a structure based piecewise linear(PWL)method to deal with the relationship between leakage and temperature and build thermal model.Combine the traditional model predictive control method to management the temperature of chip.However,this method has a problem of consumption burden for large-scale chips.Next,in response to this problem,a new predictive dynamic thermal management method with neural network thermal model is proposed to naturally consider the inherent nonlinearity between leakage and temperature,which is data based black box identification method.We start with analyzing the problems of using recurrent neural network to build the nonlinear thermal model,and point out that there is exploding gradient induced long-term dependencies problem,leading to large model prediction errors.Based on this analysis,we further propose to use echo state network(ESN),which is a special type of RNN,as the leakage-aware nonlinear thermal model.We theoretically and experimentally show that ESN achieves much higher accuracy by completely avoiding the long-term dependencies problem.On top of this nonlinear ESN thermal model,we propose a novel model predictive control scheme called ESN MPC,which uses iterative steps to find the optimal future power recommendations for thermal management.Finally,the recurrent neural network based dynamic thermal management method for multi-core systems is verified experimentally.The method in this paper outperforms the existing linear model based dynamic thermal management method in both temperature management quality and computing overhead.Being able to consider the leakage-temperature nonlinear effects and equipped with advanced control technique,the recurrent neural network based dynamic thermal management method achieves an overall high quality temperature management with smooth and accurate temperature tracking.