论文部分内容阅读
本文提出一种性能驱动的MCM划分神经学习方法.新算法具有如下特点:(1)允许功能设计和布图设计同时进行,(2)划分时,不仅考虑了模块间的逻辑关系,还考虑了MCM的版图结构.(3)具有芯片间连线数目最少和时钟周期最短双重优化目标.(4)能使连线尽可能产生在相邻近的芯片之间.(5)网络的结构合理,学习速度快.
In this paper, we propose a performance-driven method of MCM partitioning neural learning. The new algorithm has the following characteristics: (1) allows functional design and layout design to be carried out at the same time, and (2) divides not only the logical relationship among modules but also the layout structure of MCM. (3) with a minimum number of connections between the chip and the clock cycle of the shortest dual optimization goals. (4) can make the connection as much as possible between adjacent chips. (5) The structure of the network is reasonable and the learning speed is fast.