The Speedup Research And Thermal Design Of Multicore Processors

At the beginning of this century, CPU appeared the inflection point of lifting speed of relying solely on increasing the main-frequency. So far multi-core processor emerges as the times require, and becomes study of hot. Two important evaluating parameters are concerned by this work for processing speed rising and power assumption depressing through theoretical calculation.There are two main goals in this thesis:one is to recalculate the Speedup of multi-core based on Amdahl’s Law, so that the IC designer can design faster-processing processors, the other is to discuss some low-power design methods for basic thermal design of multi-core processor especially three dimensional processor.There are4main parts in this thesis:in the first part, we discuss the existing research on parallel calculation. We measure the processing speed with calculating time, which brings about the concept of Speedup.In the second part, we review the Amdahl’s Law, which is the basic standard for evaluating the parallel calculation speedup. Then we combine Rent’s Rule with Amdahl’s Law, and propose a new model to describe the speedup based on First Principle.In the third part, we compare the symmetric multi-core architecture with asymmetric architecture, and discuss the time spending description. With a precise description on asymmetric, we calculate and condition where symmetric architecture is better than symmetric one.In the fourth part, we analyze the power consumption distribution and type in an integrated circuit, and summarize the strategy and methods of low-power design. Then we discuss three ways of thermal design of multi-core:thermal vias; voltage islands; dynamic heat management. Our main results and conclusions are as following:(1) our new model on speedup is practicable after validation;(2) asymmetric multi-core architecture is better than symmetric architecture on most conditions.