Research On Task Allocation And Scheduling Method For 3D NoC With Temperature Optimization

In the multiprocessor System-on-Chip based on 3D NoC, the position of the task in the MPSoC determines the distribution of the computational and communication power consumption of the system, which has an important influence on the temperature of the chip. Task allocation and task scheduling determine the distribution of the task in the initial stage of the MPSoC operation and the process of operation, respectively. Therefore, it's very important to study the task allocation and scheduling to achieve the temperature optimization of MPSoC.The main work of this thesis was designed the temperature optimization mechanism based on task allocation, temperature information shared and task scheduling, realized temperature optimization of MPSoC based on 3D NoC. Firstly, this thesis designed a task allocation algorithm 3D-EOTA through the establishment and analysis of the communication power model and thermal model, which was considered computational and communication power consumption and heat dissipation capability of the system in different locations. Secondly, this thesis proposed a multicast routing algorithm 3D-POM which made full use of the public transmission path and characteristics of TSV, reduced communication delay and communication power consumption and thus designed of the temperature information sharing mechanism. Finally, based on the temperature information sharing mechanism, temperature prediction model and task migration strategy, this thesis designed a distributed task scheduling method, avoided the risk of communication congestion and local hot spots caused by centralized temperature management.In the simulation environment based on Noxim and HotSpot, when the 3D NoC was 4x4x3 and test case was random, compared with Coldest-First, the 3D-EOTA reduced communication power consumption and peak temperature by 52.23% and 4.31 K, respectively. Compared with MXYZ, the average communication delay of 3D-POM was optimized by 4.76%, and the communication power consumption was optimized by 15.21%. When the scale of 3D NoC was 8x8x3, compared to the traditional task scheduling algorithm, 3D-DTS reduced time of hot and number of task migration by 3.59% and 14.62%, respectively.