AN EFFICIENT ALGORITHM FOR VLSI NETWORK PARTITIONING PROBLEM USING A COST FUNCTION WITH BALANCING FACTOR

This paper presents an efficient algorithm for network partitioning pr oblem, which improves Fiduccia and Mattheyses' (F-M's) algorithm [1]. We have noticed that the main problem of F-M's algorithm is that the c ell move operation is largely influenced by the balancing constraint. In order to handle this kind of inherent limitation in F-M's algorithm , a cost function similar to the penalty function used in [121 is adop ted which reflects balance degree of a partition as well as its cutset size. The weighting factor R is introduced in the cost function to de termine the relative importances of the two factors: cutset size and b alance degree. Using this cost function, we propose an iterative impro vement algorithm which has the time complexity of O (b (m + c2)), wher e b is the number of blocks, m is the size of network, and c is the nu mber of cells. It is proven that the proposed algorithm guarantees to find a balanced partition if the value of R satisfies a certain condit ion. Experimental results show that the proposed algorithm outperforms F-M's algorithm in most cases.