The Study Of Parallel Computing And Software Design For 3D Thermal Simulation

With the development of systems on package(SoP)and manufacturing process technique,the integration density of SoP structure is increasing sharply.Consequently,the computing on thermal characterization of integrated circuits(ICs)becomes more and more complex.The traditional serial computing is becoming challenging due to the increased problem scale.Meanwhile,multi-core computers and general purpose GPU are in rapid development,parallel computing is becoming the mainstream computing method.The study of parallel computing for 3D integrated structure thermal simulation makes great sense to improve efficiency of simulation and reduce the waste of computing resources.This thesis mainly focuses on thermal effects of SoP,the parallel computing of two kinds of thermal simulation algorithms is studied.A electro-thermal simulation platform is developed.The parallel version of Alternating Direction Implicit(ADI)algorithm is implemented on multicore computer system.The accuracy and performance are validated through the transient thermal response comparison between the parallel algorithm and commercial software.Furthermore,the algorithm is applied to GPU device,to meet the coalesced requirements in GPU memory accesses,the Cyclic Reduction(CR)algorithm is introduced,the optimization for GPU memory access is implemented.Compared with single-core serial ADI computing,The GPU version ADI algorithm is 20 times faster.At last,Finite Element Method(FEM)is employed for complex structure thermal simulation,the most time-consuming part: Preconditioned Conjugate Gradients(PCG)parallelization is studied,and the GPU version of PCG is implemented.Meanwhile,the electro-thermal simulation platform based on this algorithm and HFSS is developed,It's suitable for engineering applications.