Custom hardware acceleration solutions for real-life embedded systems: Potential and challenges

Embedded systems are now increasingly used to replace typical Personal Computers (PCs) in many application areas. Major reasons behind this move are reliability, cost, scalability, and design density. Deploying embedded solutions in PC-dominated areas requires software re-design or adaptation, efficient hardware/software codesign and interface design, and further system optimization at all abstraction levels. More efficient and automated design techniques and tools are therefore required to improve the design quality of embedded systems and achieve software re-usability. This paper presents the state-of-the art techniques and selected commercial tools which can be applied for that purpose. Examples of these tools are: C2H compiler and custom instructions for soft CPUs (NIOS II) and Catapult. In this context, the operation and applicability of these tools are discussed and compared concerning the quality of the entire design. Factors which influence the process of selecting one tool and/or design technique such as the regularity and the nature of the algorithm also have to be considered. A case study is taken from a real-life application and covers the process of partitioning security Application Programming Interfaces (APIs) in hardware (FPGA) and software: data-dominated parts are efficiently implemented here in hardware whereas control-dominated parts are realized in software. Software parts are implemented using an arbitrary programming language which can be C/C++ or Java. Besides the hardware and software parts, an efficient interface is developed in between. The quality of the entire design in the case study is evaluated not only in terms of performance but also considering hardware cost and system flexibility