论文部分内容阅读
本文在提出一个直接和间接相联模块间相似性的表示方法的基础上,提出了一个基于自组织神经网络的性能驱动MCM划分的神经学习方法.算法求解如何在高层设计中将功能模块分配到MCM芯片中.算法不仅考虑了模块间的相似关系,还考虑了MCM的版图结构;具有芯片间连线数目最少和时钟周期最短双重优化目标;能使连线尽量产生在相邻近的芯片之间;能满足时延、散热和面积约束.文中还提出了一个层次神经网络模型和面积约束下的MCM划分的层次神经学习方法.本文的神经网络模型结构合理,学习速度快.
In this paper, a new method of performance-driven MCM partitioning based on self-organizing neural networks is proposed based on the representation of similarity between direct and indirect modules. The algorithm solves the problem of how to assign functional modules to MCM chips in high-level design. The algorithm takes into account not only the similarity between the modules but also the layout structure of the MCM. The algorithm has the advantages of minimizing the number of interconnections between chips and minimizing the clock cycle, and can optimize the connection between the adjacent chips as much as possible. Delay, heat and area constraints. In the paper, a hierarchical neural network model and hierarchical neural learning method for MCM partitioning under area constraints are also proposed. The neural network model in this paper has a reasonable structure and fast learning speed.