EXPLORER An Interactive Floorplanner for Design Space Exploration.pdf

EXPLORER : An Interactive Floorplanner for Design Space Exploration Henrik Esbensen Avant! Corporation 1208 East Arques Avenue Sunnyvale, CA 94086, USA esbensen@ Ernest S. Kuh Department of EECS University of California Berkeley, CA 94720, USA Abstract An interactive oorplanner based on the genetic algo- rithm is presented. Layout area, aspect ratio, routing congestion and maximum path delay are optimized si- multaneously. The design requirements are rened in- teractively as knowledge of the obtainable cost tradeos is gained and a set of feasible solutions representing al- ternative, good tradeos is generated. Experimental re- sults illustrates the special features of the approach. 1 Introduction When determining the oorplan of an integrated cir- cuit (IC) the objective is to nd a solution which is satisfactory with respect to a number of competing cri- teria. Most often specic constraints have to be met for some criteria, while for others, a good tradeo is wanted. The approach taken by virtually all existing tools is to minimize a weighted sum of some criteria subject to con- straints on others. I.e., if k criteria are considered, the objective is to minimize the scalar-valued cost function c = jX i=1 wici s. t. 8 i = j + 1; : : : ; k : ci  Ci (1) for some j, 1  j  k. Here ci is the cost of the solution with respect to the ith criterion and the wis and Cis are user-dened weights and bounds, respectively. However, at the oorplanning stage of the design pro- cess, the expected values of the cost criteria are based on relatively rough estimations only. Furthermore, the available information on obtainable tradeos, e.g. the relationship between area and delay, is very limited or non-existent. Since the notion of a \good solution in- herently depends on which tradeos are actually obtain- able, the overall design objective is rarely clearly den- able. Consequently, it may be very dicult to specify a set of weight and bound values that makes a tool based