1 / 28

A Novel Fixed-outline Floorplanner with Zero Deadspace for Hierarchical Design

A Novel Fixed-outline Floorplanner with Zero Deadspace for Hierarchical Design. Ou He, Sheqin Dong, Jinian Bian , Satoshi Goto , Chung- Kuan Cheng Tsinghua University, Beijing, China Waseda University, Kitakyushu, Japan University of California, San Diego, La Jolla, CA.

shelby
Download Presentation

A Novel Fixed-outline Floorplanner with Zero Deadspace for Hierarchical Design

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A Novel Fixed-outline Floorplanner with Zero Deadspace for Hierarchical Design Ou He, Sheqin Dong, JinianBian, Satoshi Goto, Chung-Kuan Cheng Tsinghua University, Beijing, China Waseda University, Kitakyushu, Japan University of California, San Diego, La Jolla, CA ICCAD 2008

  2. Outline • Introduction • Methodology • Experiment • Conclusions & Future Work

  3. Outline • Introduction • Methodology • Experiment • Conclusions & Future Work

  4. Introduction • Floorplanning • Two optimizations oriented: Area, Wirelength • SA is often used for solving this kind of problem • Fixed-outline floorplanning • Dominates modern hierarchical designs • Much harder than outline-free • Soft module floorplanning • Modules with a certain area but variable aspect ratios (ARs) • Achieves high area utilization (more than 99%) • Easy to implement by Linear/Quadratic Programming

  5. Introduction • Prevoius work • ZDS (Zero-Dead-Space) (Cong et al, TCAD 2006) • A deterministic method with some heuristics • Very fast • Less extendable and powerful on other optimization metrics (ex: wirelength) • Convex optimization techniques (Zhan et al, ASPDAC 2006; Luo et al, ASPDAC 2008) • Able to get a global minimum • Hard to model every metric and still remain the problem convex

  6. Introduction • Research motivation • Need not to minimize the area of floorplan • High area utilization means small area occupied • We can concentrate on other optimization metrics • Wirelength • Noise • Hot spot • Heat disspation • Etc.

  7. Outline • Introduction • Methodology • Experiment • Conclusions & Future Work

  8. Methodology • Techniques • Basic packing • Packing the modules by topological representation • Ordered Quadtree • A new topological representation • Local Refinement • Protect modules from extreme Ars • Incremental packing • Speedup the optimization

  9. Methodology • Basic Packing • Place all the modules following a given topological structure • Produces zero deadspace • Handle the fixed-outline constraint in any ARs with 100% successful rate • Builds and solves a group of four quadratic equations in four variables iteratively

  10. Methodology: Basic Packing • Step 1: Partitioning sub-regions • Get 8 sub-regions when placing Module i in a fixed-outline area • S1,S2,S3,S4 have two possible topological relations to Module i • There will be 16 cases for the assignment

  11. Methodology: Basic Packing • Step 2: Picking up one suitable assignment of the 16 cases • One tip for the picking • The largest subset of a(i), b(i), l(i) and r(i) wins the largest number of S1, S2, S3 and S4. • where,

  12. Methodology: Basic Packing • Step 3: Building the equations • A group of quadratic equations in four variables (w1, w2, h1 and h2) is built up as follows. • Finally, w1, w2, h1 and h2 will decide the coordinate and AR of Module i.

  13. Methodology: Basic Packing • Step 4: Solving the equations • There are 16 different equations corresponding to 16 possible cases. • Solve the 4 children modules in sub-regions until all the modules are packed.

  14. Methodology • Ordered Quadtree • Is custom-made for the basic packing to facilitate its integration to SA • Each node has four subtrees, which represent modules above, below, left to and right to the node respectively

  15. Methodology: Ordered Quadtree • Perturbations on Ordered Quadtree • Subtree swap (Subtree 4 and Subtree 6) • Node migration (Node 3) • Node swap (Node 2 and Node 3) • Basic packing is called after each perturbation a b c

  16. Methodology • Local Refinement • Topological Local Refinement (TLR) • Geometric Local Refinement (GLR) • Preventing modules from extreme ARs • Performed locally on a floorplan to keep an acceptable increase on the wirelength

  17. Methodology: Local Refinement • Geometric Local Refinement (GLR) • Enumerates all the possible Abnormal Modules in a local sub-region of two or three modules. • Corrects those modules directly with no operations on the original ordered quadtree.

  18. Methodology: Local Refinement • Topological Local Refinement (TLR) • Handles Abnormal Modules in a local sub-region of more than three modules • Changes the original quadtree and restart basic packing • More time consuming TLR Restart Basic Packing

  19. Methodology • Incremental Packing • Basic packing does not need to start all over again after each perturbation. • The incremental packing can be operated from the lowest common ancestor of two perturbed branches

  20. Outline • Introduction • Methodology • Experiment • Conclusions & Future Work

  21. Expriment • Chip wirelength is the only optimization metric after 100% area utilization is achieved. • Compared with two Non SA-based fixed-outline floorplanners on soft modules (AR constraint [1/3, 3]) • Zhan’s work (A-FP, ASPDAC 2006) • Cong’s work (ZDS, TCAD 2006) • Compared with three SA-based outline-free floorplanners on soft modules (AR constraint [0.1, 10]) • Kim’s work (SA-CT+LP and SA-LP, TCAD 2003) • Lin’s work (SM-FP1, ICCAD 2006) • Itoga’s work (SM-FP2, IEICE 2005)

  22. Expriment: non SA-based Table 1. Comparisons of Wirelength (μm) between A-FP (Zhan’s Work) and SAFFOA (our work)

  23. Expriment: non SA-based Table 1(cont’d). Comparisons of wirelength (μm) between A-FP (Zhan’s Work) and SAFFOA (our work)

  24. Expriment: non SA-based Table 2. Comparisons between ZDS (Cong’s work) and SAFFOA (our work)

  25. Expriment: SA-based Table 3. Experimental results with SM-FP *The experimental results are updated with the permission from the author.

  26. Expriment: SA-based Table 3(cont’d). Experimental results with SM-FP

  27. Outline • Introduction • Methodology • Experiment • Conclusions & Future Work

  28. Conclusions & Future Work • Conclusions • The FIRST SA-based Fixed-outline Floorplanner with the Optimal Area utilization named SAFFOA is implemented. • Future Work • A better local refinement will bring a great increase on the efficiency and reliability • Mixed-mode SAFFOA and 3D SAFFOA

More Related