Effective regularity extraction and placement techniques for datapath-intensive circuits

摘要

Regular structures, like datapath, are important components of integrated circuits. Datapath logic is usually placed with high regularity and compactness for higher performance by using manual placement. The authors propose effective datapath regularity extraction and placement (DREP) techniques which simultaneously place datapath logic and random logic. This method detects datapath logic and effectively formats regular datapath structures while optimizing the order of functional stages and placement of datapath blocks. Moreover, the datapath structures are further optimized by using bit-slice order adjustment and partitioning techniques during global placement. Partitioning of a big datapath macro greatly increases the placement flexibility, since partitioned sub-blocks of the datapath macro can be optimally placed with other blocks. A new effective method is also suggested to decide the block to be partitioned and the granularity of partitioning. Similar to the manual placement results, the datapath logic is regularly placed and the datapath cells are aligned well, vertically or horizontally by the DREP techniques. When compared with the state-of-the-art works, the experimental results show that the new techniques produce significantly better results than other methods in terms of half perimeter wire length, Steiner wire length, and routability, measured from the detail routing results.