1. Configurable Multi-product Floorplanning Qiang Ma, Martin D.F. Wong, Kai-Yuan Chao ASP-DAC 2010

2. Outline Introduction Problem Formulation Methodology  Exact Algorithm  Greedy Heuristic Experimental Results Conclusions

3. Introduction Problem Formulation Methodology  Exact Algorithm  Greedy Heuristic Experimental Results Conclusions

4. Introduction Conventional ASIC or SoC design floorplan usually targets for one single product; and, high efforts in re-floorplan and re- convergence for different products are still required if there is no pre-design stage multi- product planning Therefore, the problem of designing floorplans at product or market planning stage that simultaneously optimizes multiple products

5. Introduction Traditional floorplanning problem  Given a set of rectangles, each with a range of aspect ratio  Generate a packing of all the rectangles s.t. The rectangles do not overlap The total area and wire length are minimized

6. Introduction Multi-product Floorplanning  Generate a floorplan of enough components (basic blocks) that is configurable for all the products  The rectangular region for each product is as small as possible

7. Introduction Problem Formulation Methodology  Exact Algorithm  Greedy Heuristic Experimental Results Conclusions

8. Notations Problem input  {b 1, b 2, …, b m } be the set of m basic blocks  {P 1, P 2, …, P q } be the set of q products  [D i 1, D i 2, …, D i m ] T be the Demand Vector D i of P i

9. Notations Cutline  a line obtained by extending the interval of a block boundary Valid Region  A rectangular region R on a candidate floorplan formed by four Cutlines l, r, t, b (denoted by R(l, r, t, b)) Capacity Vector  C R = [C R 1, C R 2, …, C R m ] T of a Valid Region R, where C R j denotes # of basic block b j lying within R Feasible Region  R is Feasible for product P i if and only if C R ≧ D i

10. Problem Formulation Multi-product Floorplanning (MPF)  Inputs: {b 1, b 2, …, b m } {P 1, P 2, …, P q } Each product P i is associated with a Demand-Vector D i  Objective: Generate a floorplan of sufficient basic blocks to accommodate all the q products The total area of the MPF for each product on the resultant floorplan is minimized

11. Introduction Problem Formulation Methodology  Exact Algorithm  Greedy Heuristic Experimental Results Conclusions

12. Preprocessing

13. Exact Algorithm

14. Horizontal Scan (H-Scan)  Determine a Horizontal Feasible Regions (HFR) for a product  Horizontal interval array H is scanned  Fix left boundary, shift right boundary until the current region contains sufficient basic blocks

15. Exact Algorithm Horizontal Scan (H-Scan)

16. Exact Algorithm Vertical Scan (V-Scan)  Given a HFR(lb, rb), find HFR*(lb, rb) with minimum height  Find positions for tb and bb s.t. (bb-tb) is minimum  At the beginning, tb = Top-Cutlines[0], bb = Bottom-Cutlines[0]  Repeat Shift bb down until R(lb, rb, tb, bb) is feasible Shift tb down until R(lb, rb, tb, bb) is infeasible Record the current height and update the best  Until bb is no longer able to be shifted downward

17. Exact Algorithm Vertical Scan (V-Scan)

18. Greedy Heuristic Given a candidate floorplan and a product P i ’s demand vector D i, the greedy heuristic behaves as follows  The feasible region R(lb, rb, tb, bb) starts with the whole floorplan  Greedily shrink the four boundaries Pick one boundary b and shift it one step towards the center Its shifting reduce R by the most  Break the tie by picking the least recently used boundary The feasibility of R is maintained Scan Horizontal interval array H and vertical interval array V Stop when R can not be shrunk any more The resultant region R is a Minimal Feasible Region

19. Greedy Heuristic

20. Introduction Problem Formulation Methodology  Exact Algorithm  Greedy Heuristic Experimental Results Conclusions

21. Experimental Results Exact  Use the exact algorithm Greedy  Use the greedy heuristic Hybrid  Applying the greedy heuristic at the first stage and then switching to the exact algorithm at the second stage, when the acceptance ratio drops below a certain value  Provide good tradeoff between solution quality and rum time

22. Experimental Results The three method are tested and compared on a set of test cases derived from industrial data

23. Experimental Results

26. Introduction Problem Formulation Methodology  Exact Algorithm  Greedy Heuristic Experimental Results Conclusions

27. This paper introduced the Multi-product Floorplanning problem  Simultaneously design for a family of related products  Significantly reduce overall design time and cost First work in literature addressing this problem The effectiveness of this approach is validated by promising results on a set of test cases