GSRC  Bookshelf (see other slots )

ISPD06 Floorplacement Benchmarks

Aaron Ng, Rajat Aggarwal, Venky Ramachandran, Igor Markov

Contents

I.  Introduction 
II.  Calypto Benchmarks
III.  IBM-HB+ Benchmarks
IV
 Literature
I. Introduction

This entry includes floorplacement benchmarks for CAD research. There are 2 suites of benchmarks. The Calypto suite are floorplacement benchmarks derived from industrial netlists. The IBM-HB+ suite are benchmarks created by modifying the IBM-HB benchmarks prepared at UCLA. The IBM-HB+ benchmarks have a greater variation in macro sizes, while maintaining the same total cell area. Both suites have no fixed modules, except for peripheral I/O pads. More information about these benchmarks and empirical results are available in [1].

II. Calypto Benchmarks
This suite of benchmarks is made available by Calypto Design Systems, Inc. and are derived from customer chips (e.g., CPUs and video ICs) and internally-generated regression tests. The size of the benchmarks ranges from 81 to 8827 RTL modules. The benchmarks consist primarily of movable macros, but some benchmarks also contain standard cells. All the macros in the benchmarks are soft (have variable aspect ratios). This suite of benchmarks was originally problematic for various reasons: i) there can be a large variation of size between macros, ii) there can be a large number of movable macros, and iii) there can be a combination of soft/hard macros and standard cells.
Characteristics of the proprietary designs
Proprietary designs Movable modules Nets Arealargest
(%)		 Arealargest /
Areasmallest
Cells Macros
cal040 1 4605 4607 0.1 650
cal098 3200 1212 4673 0.1 529
cal336 17 105 147 2.2 11556
cal353 217 459 908 7.0 11556
cal523 934 1936 4350 0.3 3080
cal542 7 74 92 20.1 11556
cal566 93 1553 5502 1.2 11556
cal583 773 1530 3390 0.4 2916
cal588 293 495 1111 0.6 900
cal643 139 316 598 6.5 6162
calDCT 0 8827 11463 50.0 185330

cal040
cal098
cal336

cal353
cal523
cal542

cal566
cal583
cal588

cal643
calDCT
The 11 benchmarks available in Bookshelf format: Calypto_ISPD06.tar.gz
The Calypto benchmarks have been updated as of August 2, 2006. Please download them again if you have an older version.

Disclaimer: If you are using these benchmarks in a publication or a technical report, please cite our publication [1] in addition to the URL. It describes the origin of the benchmarks.

III. IBM-HB+ Benchmarks

These benchmarks were derived from the IBM-HB benchmarks. The benchmarks range from 500 to 2000 blocks in size and contain both soft and hard macros. The goal of the modifications to the original IBM-HB benchmarks was to create floorplacement instances that contain the features resembling those found in the Calypto designs. To reproduce difficult instances, each benchmark in the original IBM-HB suite was modified as follows: the largest macro was inflated by 100% while the areas of the remaining macros were reduced to preserve the total cell area. Note that a more detailed study can involve varying the dimensions of more than one macro at a time, and even more difficult floorplacement problems can be constructed.

Characteristics of the IBM-HB+ benchmarks
Benchmarks Movable modules Nets Arealargest
(%)		 Arealargest /
Areasmallest
Cells Macros
ibm-HB+01 0 911 5829 6.4 8416
ibm-HB+02 0 1471 8508 11.3 3004.3
ibm-HB+03 0 1289 10279 10.8 33088
ibm-HB+04 0 1584 12456 9.2 13296.5
ibm-HB+06 0 749 9963 13.6 18173.8
ibm-HB+07 0 1120 15047 4.8 399.5
ibm-HB+08 0 1269 16075 12.1 50880
ibm-HB+09 0 1113 18913 5.4 29707
ibm-HB+10 0 1595 27508 4.8 71299
ibm-HB+11 0 1497 27477 4.5 9902.3
ibm-HB+12 0 1233 26320 6.4 74256
ibm-HB+13 0 954 27011 4.2 33088
ibm-HB+14 0 1635 43062 2.0 17860
ibm-HB+15 0 1412 52779 11.0 62781.3
ibm-HB+16 0 1091 47821 1.9 31093
ibm-HB+17 0 1442 56517 0.9 12441
ibm-HB+18 0 943 42200 1.0 3384

ibm01
ibm03
ibm04

ibm06
ibm08
ibm10

ibm12
ibm14
ibm16

ibm18

The IBM-HB+ benchmarks have been updated as of August 2, 2006. Please download them again if you have an older version.
The 17 benchmarks, available in Bookshelf format: IBM-HB+_ISPD06.tar.gz

Disclaimer: If you are using these benchmarks in a publication or a technical report, please cite our publication [1] in addition to the URL. It describes how the benchmarks were created.

IV. Literature

[1] A. N. Ng, R. Aggarwal, V. Ramachandran and I. L. Markov, "Solving Hard Instances of Floorplacement", International Symposium on Physical Design (ISPD), San Jose, 2006.