A systematic software, firmware, and hardware codesign methodology for digital signal processing

Abstract

Creating an embedded system that meets its functional, performance, cost, and schedule goals is a software-and-hardware codesign problem, since the design of the software and hardware components influence each other. The traditional design methodology is sequential, with hardware designed first and then software. The lack of a unified and unbiased approach can lead to suboptimal design and incompatibilities across the software and hardware boundary. To solve these problems, we propose a new software/firmware/hardware codesign methodology to systematically build correct designs efficiently. This codesign methodology includes requirements development, architecture forming, software/ firmware/hardware partitioning, design-pattern mapping, new-design pattern synthesis, integration, and testing. We tested our methods on three application areas. One was a digitizer-filter architecture for ultra-high frequency signals for which we synthesized design patterns in firmware to meet high-frequency requirements. Another was a digitizer-filter architecture for low-frequency signals. A third was a hidden Markov model using dynamic programming. We implemented and tested the first application on a Tektronix/Synopsys embedded system and the second on a Pentek embedded system based on the requirements provided by the stakeholders