Year: 2020
Type: dissertation
Abstract: Coarse-grained reconfigurable architectures (CGRAs) provide high energy efficiency with word-level programmability rather than bit-level ones such as Field Programmable Gate Arrays (FPGAs).Therefore, the CGRAs are expected to be used for embedded systems, IoT (Internet of Things) devices, and edgecomputing.In essence, a CGRA is an array of numerous processing elements (PEs).In order to exploit this abundant computation resource, a compiler for CGRAs has to fulfill more tasks compared that for general-purpose processors.Thus, many studies have proposed optimization methods, especially for application mapping, because the performance and energy efficiency strongly depend on optimization at compile time.However, most work focuses only on performance improvement or resource minimization, although such optimization objectives are not always appropriate when considering various use cases.In this thesis, an application mapping framework using multi-objective optimization based on a genetic algorithm (GenMap) is proposed.The proposed framework does not depend on any specific architecture and can consider various optimization objectives.Thus, users can easily customize fitness functions for optimization.Firstly, this thesis defines the mapping problem to be addressed by the proposed method.Then, a novel formulation of this problem for the genetic algorithm is presented.However, if the whole of the problem is solved by only the genetic algorithm, it needs prohibitive optimization time to obtain reasonable solutions.Therefore, several heuristics are also proposed and integrated into the framework.Furthermore, this thesis provides aggressive power optimization methods based on a dynamic power model and integer linear program (ILP) formulation for leakage minimization.Three fabricated CGRA chips are evaluated using the proposed framework.Experimental results show that 15.7% of the wire length is reduced while keeping PE utilization compared to conventional methods.In addition, according to real chip experiments, 12.1-46.8% of energy consumption is reduced, and up to 2x speed-up is archived for several architectures compared to the other two approaches.
Authors: コジマ, タクヤ, Kojima, Takuya, 小島, 拓也
Institution: Keio University
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7139652761
