Improved Two-Way Min-Cut Circuit Partitioning with Complex Resource Constraints Using Dynamic Locking and Node Replication Techniques

• Main Authors: Ming-Shiun Lin, 林明勳
• Other Authors: Ting-Chi Wang
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/73334527986113356345

碩士 === 中原大學 === 資訊工程研究所 === 89 === Circuit partitioning has played a very important role at different levels of VLSI design. Recently, a new circuit partitioning problem, called two-way min-cut partitioning for FPGAs with complex resources, was addressed by Liu, Zhu and Wong [18]. For this type of FPGAs, multiple types of resources are contained. Each resource type has multiple choices to implement a basic cell in the circuit, and each basic cell also has multiple choices to be implemented by a resource type. Given a circuit G and two resource sets R1 and R2, the two-way min-cut partitioning problem with complex resource constraints is to partition G into two subcircuits V1 and V2 such that the total number of cut nets between V1 and V2 is as small as possible, subject to the constraints that the nodes in V1 can be implemented by the resources in R1, and V2 can be implemented by the resources in R2. To solve the problem, we study in this thesis how to apply the dynamic locking and node replication techniques to improve a given initial partitioning, respectively. In particular, we have obtained the following three significant and encouraging results: (1)We have modified the FFC-fm algorithm [18] by using the dynamic locking scheme. The experimental results show that this modification generated better cut size quality than the FFC-fm algorithm by 8.07%~75.22%. (2)We have developed an efficient algorithm for node replication. The experimental results show that using the solution generated by the FFC-fm algorithm as the initial solution, the replication algorithm could further reduce the cut size quality by 23.05%~83.57%. (3)Finally, we combined both dynamic locking and node replication, and the experimental results show that the cut size could be improved by 35.67%~99.39% as compared to the FFC-fm algorithm.