Font Size: a A A

Research On Distributed Thermal Modeling And Thermal Management Technology For Many-core Systems

Posted on:2019-07-27Degree:MasterType:Thesis
Country:ChinaCandidate:L XuFull Text:PDF
GTID:2348330569487856Subject:Microelectronics and Solid State Electronics
Abstract/Summary:PDF Full Text Request
In order to solve the problem of thermal reliability of many-core chips,the research of thermal modeling and thermal management technology is becoming more and more active.Compared to the chip with less core number,the difficulty of the thermal management technology of the many-core system is that the thermal management overhead caused by the large size of the chip thermal model is too large.In this paper,the research topic is distributed thermal modeling and thermal management,and we focus on building a distributed thermal model and a distributed model predictive control method for many-core systems in the various stages of thermal management.The main contents include the following three aspects.For distributed thermal modeling,aiming at the difficulties in dealing with heat exchange between local modules,a dual solution of boundary buffer layer and boundary equivalent heat source is put forward.In the chip thermal system packaging modeling,based on the traditional lumped thermal modeling architecture,we set the idea of HotSpot compact thermal modeling,and keep the main heat dissipation path of chip packaging unblocked while reducing the size of the model.In the packaging modeling,based on the traditional lumped thermal modeling architecture,combined with the idea of HotSpot compact thermal modeling,we keep the main heat dissipation path of chip packaging unimpeded,meanwhile the size of the model is reduced.Finally,the verification method of distributed thermal model and its application in the thermal management are explained.For distributed model predictive control,the difficulties of its design are studied.We foucus on that local module controller must know the adjacent core temperature and the future control strategy of adjacent modules.In order to solve the two difficulties,a new distributed thermal model considering the adjacent core temperature from our new modelling method mentioned above is adopted,and the iterative algorithm is used to calculate the prediction temperature of adjacent modules as the information of future control strategy of adjacent modules.Finally,the above two measures are applied to the distributed model predictive control,which is presented by a new state space equation.Finally,the distributed thermal modeling and distributed model predictive control proposed in this paper are verified experimentally.It is compared with the existing distributed thermal modeling and distributed model predictive control schemes based on the many-core systems.The experimental results show that the size of the thermal model can be effectively reduced by our modelling mothd,and the distributed thermal model built by our method have the high accuracy.Further,the distributed thermal modelling proposed by this paper support a variety of divion for local module,therefore it has some flexibility.The verification results of the distributed model predictive control method show that the our new method has high control accuracy and control effect,so it can provide an accurate control strategy for distributed thermal management of many-core chips.
Keywords/Search Tags:many-core system, thermal management, distributed thermal modeling, distributed model predictive control
PDF Full Text Request
Related items