| HW/SW co-design has unified HW/SW representations before HW/SW partitioning. In the following design processing, Systems are divided into different sub-modules before HW/SW partitioning. This procedure is called a granularity generation procedure. and a sub-module is a granularity. These granularities will be seen as basic units to be mapped to hardware or software components. The size and construction of a granularity have very flexible choices. So it has great impact on the overall system performance and cost.In this dissertation, the function relation between granularity variables and system cost is found out. It is a discrete function, but it is a approximate continuous function. Simulated by a continuous function, the origin function's character can be kept. When explore the design space, a lot of data can be got, by these data, the function can be got too. In the following design space exploring procedure, this function can result in a reasonable granularity for the design. Algorithm presented here proves a self-adaptable flexible granularity approach for HW/SW co-design is presented. Theory and algorithm models are established for generation, evaluation, selection, and adjusting of granularities. All experiment results are laid out and analyzed in detail.The algorithm can resolve the question of how to adjust a design granularity during the design procedure. It allows to consider the system cost during granularity selection stage, and it allows granularities with different size exist in one design. As can be seen in this dissertation, the system cost implemented by this algorithm will be much better than other design granularity approaches.By this self adaptable granularity approach to complex system design, the designer's intervening can be reduced at front end of SOC design, more choices can be obtained to implement a system using different architecture. As a result, different styles of synthesize and optimization can be met at back end design.
|
PDF Full Text Request