Directional and Single-Driver Wires in FPGA Interconnect

1. Directional and Single-Driver Wiresin FPGA Interconnect Guy Lemieux Edmund Lee Marvin Tom Anthony Yu Dept. of ECE, University of British Columbia Vancouver, BC, Canada

2. Outline • Motivation • Bidirection vs. Directional • New detailed routing architecture • Which is better? • Tristate vs. Single-Driver • HSPICE results • Which is better? • Place & Route Results • Conclusions

3. Motivation: Bidirectional Wires Interconnect Logic

4. Motivation: Bidirectional Wires Problem Half of TristateBuffers LeftUnused Buffers Dominate Size of Device

5. Bidirectional vs. Directional Wiring

6. Bidirectional vs Directional

7. Bidirectional vs Directional

8. Bidirectional vs Directional

9. Bidirectional vs Directional

10. 2-Dimensional FPGAaka Island-Style or Mesh

11. Bidirectional Switch Block

12. Directional Switch Block

13. Bidirectional vs Directional Switch Block:Directional has Half as Many Switch Elements Switch Element:Same Quantity and Type of Circuit Elements(twice the wiring)

14. Building up Long WiresStart with One Switch Element Wire ends for straight connections.

15. Building up Long WiresConnect MUX Inputs Extend MUX inputs

16. Building up Long WiresConnect MUX Inputs TURN UP from wire-ends to mux

17. Building up Long WiresConnect MUX Inputs TURN DOWN from wire-ends to mux

18. Building up Long WiresAdd +2 More Wires (4 total) Add LONG WIRES, turning UP and DOWN.

19. Building up Long WiresAdd +2 More Wires (6 total) Add LONG WIRES, turning UP and DOWN

20. Building up Long WiresSingle Layout Tile !!! Add wire twisting

21. Long Wires! 1 2 3 NOTICE: One switch element holds 6 wires #Wires := WireLength x NumDirections = 3 x 2 = 6 No “partial” switch elements with fewer wires

22. Small Switch BlockOne L3 Switch Element

23. Bigger Switch BlockTwo L3 Switch Elements NOTICE Switch element design forces quantization of channel width BidirectionalOne quantum = 1*L DirectionalOne quantum = 2*L

24. Summary • Directional wiring • Good • Potential area savings • Bad • Big input muxes, slower • Bigger quantum size (2*L) • Detailed-routing architecture is different(need new switch block) • Need to evaluate!

25. Tristate vs. Single-driver Wiring

26. Bidirectional WiringOutputs are Tristates Fanout increasesdelay Multi-driver Wiring!!! Bidir Architecture

27. Directional WiringOutputs can be Tristates Fanout increasesdelay Multi-driver Wiring!!! Dir-Tri Architecture

28. Directional WiringOutputs can use switch block muxes New connectivityconstraint Single-driver Wiring!!! Dir Architecture

29. Tristate HSPICE Model

30. Tristate HSPICE Model

31. Single-driver HSPICE Model

32. HSPICE DelaysIncludes Switch + Wire TSMC 0.18um

33. AREA * HSPICE Delay TSMC 0.18um

34. Summary • Single-driver wiring • Good • Same delay as tristate • No delay increases caused by fanout • Fewer wire loads: 27% lower capacitance • Bad • Directional only (by necessity) • Area-delay product “seems” worse, but isn’t

35. Place and Route Results

36. Channel Width

37. Area (Transistor Count)

38. Delay

39. Area-Delay Product

40. Results Summary • Average improvements usingsingle-driver wiring 0% channel width 9% delay 14% tile length of physical layout 25% transistor count 32% area-delay product 37% wiring capacitance

41. Conclusions • No more tristates! • Eliminates need for pass transistors • No “Vt” loss signal degradation • Better signal reliability, better drive strength • Significant savings in all metrics • Any reasons left to use bidirectional wiring ??? • Savings INCREASES with circuit size • Because interconnect dominates big circuits