Fast Placer Implementation (FPI) Framework

·         Fast Placer Implementation Framework is a 3-stage placer flow targeting on large scale standard cell placement problem. It first reduces the size of original placement problem by Fine Granularity Clustering (FGC). Then, in stage 2, it calls an existing fixed-die standard cell placer to perform global placement. Stage 3 is unclustering and refinement. In this flow, since the placer in the middle layer works on the reduced netlist, its efficiency can be significantly improved.

Fig 1: Fast Placer Implementation (FPI)

·         Latest experimental results with Capo 8.6

·         Experimental Results with latest Dragon. Coming up!

·         IBM-Place1.0-nc benchmarks are created from IBM-Place1.0 benchmarks with channels between core rows removed. Each benchmark has 15% ~ 18% whitespace. Click here to download IBM-Place1.0-nc benchmarks.

·         Click here to download the latest FPI version 2.0 for linux.

·         Click here for the latest Capo release.

·         Click here for the latest Dragon release.

·         Here is the FPI manual, which includes all the information you need to run fpi2.0 on your machine. I have two standard scripts script.wm and script.wm.dragon here.

·         Known limitations: FPI2.0 handles tall cells better than fat ones.

·         Related Publications

B. Hu and M. Marek-Sadowska, "Fine-granularity clustering for large-scale placement problem", Proc. ISPD 2003.

B. Hu and M. Marek-Sadowska, "Wire-length Prediction based Clustering and its Application to Placement Problem". Proc. Design Automation Conference, 2003

Bo Hu, ECE Department, University of California, Santa Barbara, CA 93117. Lab: (805)-893-5678. Email: hb@ece.ucsb.edu