A DSP-enhanced 32-bit embedded microprocessor

Reviewer: Junqing Sun

High-performance computer architectures for specific applications are widely discussed. Applications such as digital signal processing (DSP), climate simulation, biology computation, and structural analysis can require tremendous computation efforts, from embedded systems to supercomputers. As computation time is crucial to the success and cost of these important projects, higher performance architectures for these applications are as important as they are for general-purpose computers. There are multiple alternatives to achieve this goal. For example, field-programmable gate arrays (FPGAs), graphics processing units (GPUs), and cell processors are usually adopted as coprocessors that work with the central processing units (CPUs) to accelerate the computation. Normally, these architectures require special programming abilities. For example, FPGA accelerator designers need to have good knowledge of hardware and design-specific processor architectures, for the best performance of certain applications. This paper targets DSP applications, by extending the instruction set of embedded processors. The great thing about it is that the users can take advantage of this high-performance DSP unit by simply using a DSP instruction. The results show significant improvement, compared to traditional processors. This is a very well-written paper, with very clear figures. Kim and Oh provide examples for their design. Online Computing Reviews Service