Non-slicing Floorplanning-based Crosstalk Reduction on Gridless Track Assignment for a Gridless routing System with Fast Pseudo-Tile Extraction Yu-Ning Chang Yih-Lang Li Wei-Tin Lin Wen-Nai Cheng Dept of CS, National Chiao Tung University, Hsinchu, Taiwan 1
Outline � Introduction � Crosstalk-Driven GTA � Detailed Routing � Experimental Results � Conclusion 2
Why Gridless and Crosstalk- Driven Router ? � Crosstalk induced by adjacent wires increases the coupling capacitance and delay of the wires. � In modern designs, variable-width and variable-space routing is commonly used for the purpose of crosstalk and delay optimization � Gridless routers are more flexible for variable-rule routing than grid-based routers 3
4 Pseudo pin IRoute 3 IRoute 1 Preliminaries IRoute 2 Pin
sub-panel 5 Vertical panel i3 i5 i7 i2 Preliminaries i6 i4 i1 Horizontal Cut line panel
Preliminaries � Crosstalk Model � f is switching factor � l is coupling length � d is distance l = α ⋅ ⋅ i , j C ( i , j ) f β c i , j d i , j 6
Routing Flow Global router Detailed routing Gridless Global paths Detailed routing Gridless track assignment(GTA) Pattern routing GTA Preprocess: IRoute Extraction Detailed routing preprocess: Routing Initial GTA tree construction Crosstalk reduction Extended O-tree Track assignment Sub-panel Floorplan Based results re-arrangement Assignment refinement 7
Contributions of this work � This work develops a complete crosstalk- driven three-stage gridless routing system. � A congestion-driven global router � A crosstalk-driven gridless TA (GTA) � Enhanced NEMO with fast PMT extraction for detailed router. 8
Outline � Introduction � Crosstalk-Driven GTA � Detailed Routing � Experimental Results � Conclusion 9
Crosstalk-Driven GTA Gridless track assignment GTA Preprocess: IRoute Extraction Initial GTA Crosstalk reduction Extended O-tree Sub-panel Floorplan Based re-arrangement Assignment refinement 10
11 15 2/13 14 12 11 10 9 1 3 7 5 2 6 4 8 Initial GTA 16 18 17 19
Crosstalk-Driven GTA Gridless track assignment GTA Preprocess: IRoute Extraction Initial GTA Crosstalk reduction Extended O-tree Sub-panel Floorplan Based re-arrangement Assignment refinement 12
Extended O-Tree Based Assignment Refinement (EOBAR) 3 external insertion location 1 2 4 5 6 7 1 3 2 1 2 3 (a) 5 4 5 4 1 2 3 (b) 7 6 5 4 7 6 r 7 6 r Extended O-tree internal insertion location Overlap graph O-tree (d) (b) (c) 13
3 14 6 4 2 r 1 5 Extended O-Tree Based 3 Assignment Refinement 6 4 2 5 1 3 4 6 2 r 7 5 1 3 6 4 2 7 5 1
15 3 3 4 6 2 4 6 2 7 r 7 r 1 5 1 5 (d) (b) 3 Extended O-Tree Based Assignment Refinement 6 3 6 4 4 2 2 7 5 7 1 5 1 3 4 6 3 4 6 2 r Infeasible 2 r 7 1 5 1 7 5 (c) 3 (a) 6 4 3 6 4 2 2 7 5 1 7 5 1
Extended O-Tree Based Assignment Refinement � Insertion is realized by tentative plow � IRT: IRoute under test IL: Insertion Location Plowing direction: IL A IRT IRT IRT C B IRT C B IRT B IRT B IRT IL D IRT A A A IL Type II: plow IRoute Type III: plow effect propagation Type I: plow IRT path overlap to IRT to neighboring path 16
Crosstalk-Driven GTA Gridless track assignment GTA Preprocess: IRoute Extraction Initial GTA Crosstalk reduction Extended O-tree Sub-panel Floorplan Based re-arrangement Assignment refinement 17
12 12 4 3 3 4 18 7 37 13 13 20 20 22 20 14 14 14 34 Sub-panel Rearrangement 2 2 1 1 14 HIR1 HIR4 HIR3 HIR2 HIR1 HIR2 HIR3 HIR4 9 9 4 4 2 2 11 11 3 3 10 12 10 12 5 5 8 8 7 7 6 6 1 1
Outline � Introduction � Crosstalk-Driven GTA � Detailed Routing � Experimental Results � Conclusion 19
Detailed routing Track assignment results Detailed routing Detailed routing preprocess: Routing tree construction Pattern routing Gridless Detailed routing 20
21 Routing Tree Construction Pseudo pin IRoute 3 IRoute 1 IRoute 2 Pin
22 Routing Tree Construction Pseudo pin IRoute 3 IRoute 1 IRoute 2 Pin
Detailed routing Track assignment results Detailed routing Detailed routing preprocess: Routing tree construction Pattern routing Gridless Detailed routing 23
Gridless Detailed Routing 4 7 T Z 7 3 3 2 4 PMT4 PMT7 PMT3 P3 P2 PMT6 3 2 4 6 7 3 Y PMT2 2 P1 1 1 2 PMT1 1 S 1 2 5 5 5 PMT5 X A B C D 24
Bin-Based Data Structure and Fast PMT Extraction p 1 p 2 s t Un- Wire Global Pseudo Blockage routable Block path Block Region Bin-based data structure 25
Experimental Results � All routing cases were conducted on a 1.2GHz Sun Blade-2000 workstation with 2GB RAM � Six MCNC benchmark circuits using three routing layers 26
Experimental Results Table 1. Comparison of routing performance between NEMO and this work. NEMO This work Fast PMT extraction Time Mem Time Mem (Tn:sec) (MB) (Tt:sec) (MB) s5378 2.4 10 2.10 22 s9234 1.7 9 1.45 21 s13207 6.6 15 4.43 25 s15850 8.8 18 6.51 27 s38417 37.2 48 13.46 37 s38584 73.7 66 30.52 45 Tn/Tt 1.72 1 27
Experimental Results Table 2. Statistics of crosstalk reduction for fixed- and variable-rule routings. Circuit Initial assignment O-tree based refinement + HIR re- arrangement Coupling cap. x Coupling cap. x RR (%) 10 3 (C1) 10 3 (C2) FR VR FR VR FR VR S5378 .168 .123 .075 .091 55 26 .107 .040 63 S9234 .086 .047 46 S13207 .379 .294 .160 .205 58 30 S15850 .493 .363 .205 .275 59 24 S38417 1.013 .794 .385 .559 62 30 S38584 1.402 1.026 .578 .863 59 16 Ave. 59 29 28
Experimental Results • Table 3. Comparison of routing results between this work and [13]. Global Routing + NEMO Global Routeing + crosstalk driven GTA + fast PMT extraction NEMO W.L. ( × 10 4 μ m ) W.L. ( × 10 4 μ m ) Run time (sec) Run time (sec) FR VR FR VR FR VR FR VR 2.4 7.4 1.70 8.0 s5378 3.74 7.6 2.31 8.2 s9234 1.7 2.69 5.5 5.6 1.25 1.72 6.0 6.1 s13207 6.6 11.28 17 18 4.13 7.62 19 19 s15850 8.8 16.33 22 22 5.34 8.33 23 24 s38417 37.2 56.36 48 49 14.04 15.75 52 52 73.7 67 22.65 72 s38584 143.39 68 27.64 72 Comp. 2.66 4.17 0.93 0.94 1 1 1 1 29
Experimental Results Table 4. Comparison of fixed-rule routing results of a commercial routing tool and this work. Wire length ( × 10 4 μ m) circuit Run time (sec) Coupling capacitance (pf) This work CR wt SI This work CR wt SI This work CR wt SI s5378 1.70 16 8.0 7.7 3.88 4.67 s9234 1.25 14 6.0 5.7 2.28 2.72 s13207 4.13 43 19 18 8.38 9.74 s15850 5.34 49 23 23 11.29 13.77 s38417 14.04 110 52 50 21.78 23.68 s38584 22.65 157 72 69 32.69 37.06 Comp. 1 7.92 1 0.96 1 1.17 30
Conclusion � This work presents a three-stage gridless routing system � Experimental results reveal that the proposed gridless routing system can perform over 2.66 times faster than NEMO. � As compared with a commercial routing tool, this work yields an average runtime speedup of 7.92 times and an average 15% reduction rate in coupling capacitance 31
32 Thank you
Recommend
More recommend
