A Novel Methodology for the Design of Processing Elements in Embedded SIMD Architectures for Multimedia

摘要

The past 25 years have brought unprecedented growth in the performance and integration of computing systems. Recent growth in multimedia applications has created new demand for networked portable multimedia products, steering attention towards architectural approaches capable of delivering performance more efficiently. While performance remains an important aspect for computing cores targeted at portable multimedia systems, it must be combined with minimum energy consumption for maximum battery life. This paper presents a novel method to design and evaluate the architecture of processing elements for embedded SIMD architectures specialized for multimedia. Differently from other approaches in literature this methodology will combine information from the execution of an extensive application suite with detailed technology projections of important system parameters. Results will show the impact of processing element granularity in terms of throughput, energy efficiency, and area efficiency. Architectural choices will be identified that are more promising to obtain the most efficient vehicle to deliver the required performance within the constraints set by technology, cost, and battery life.