Exploration and implementation of a highly efficient processor element for multimedia and signal processing domains

摘要

The exploration and design process of highly efficient processor element for multimedia and signal processing domains is presented in this study. With the introduction of synchronous data-transfer architecture for high-performance embedded applications, the effectively exploring the exponential-size architectural design spaces by detailed simulation is intractable. The authors attack this via an automated approach. At first, its cost model is built to achieve fast and accurate estimation with the characteristic of scalability. Then, the hierarchical design space exploration methodology involving heuristic-based local process and analytical global optimisation step is proposed to achieve the approximate optimum with short time-to-market. For target domains, our proposed method arrives at better optimised results within only 25 h when compared with other methods. A System on Chip (SoC) involving the optimised processor element has been implemented in 0.13 m complementary metal oxide semiconductor (CMOS) process and the experimental results show that our processor element outperforms TMS320C64 series and does the obvious acceleration to multimedia applications in SoC system.