性能驱动系统划分的均场退火方法

Mean Field Annealing Approach for Performance Driven System Partitioning

本文通过把时延约束条件转化为一个布尔矩阵，利用这个布尔矩阵构造能量函数的时延约束项，从而解决了用神经网络处理时延约束这类不等式约束的难题，并据此提出了一个性能驱动的ＶＬＳＩ系统划分的均场退火算法．算法不仅考虑了模块间的连接关系，还考虑了版图的物理结构，是一种逻辑与版图相结合的划分方法．实验表明。

Abstract Timing constraints are transformed to a Boolean matrix used to add the timing constraints to the energy function, then the difficult problem—the inequality constraints as timing constraints treated by neural network—is solved. Based on the energy function, we propose a mean field annealing neural network approach for performance driven system partitioning. In the algorithm, not only the connections between modules but also the physical structure are considered. The results show that it has more possibility to find optimal solutions.