1. BOB-Router: A New Buffering-Aware Global Router with Over-the-Block Routing Resources Yilin Zhang1, Salim Chowdhury2 and David Z. Pan1 1 Department of ECE, University of Texas at Austin, Austin, TX, USA 2 Oracle, Austin, TX, USA ASPD’14

2. outline INTRODUCTION PRELIMINARIES BOB-ROUTER ALGORITHMS EXPERIMENTAL RESULTS CONCLUSION

3. INTRODUCTION Routing is one of the most important stages regarding performance of chip interconnection. Using IP-blocks to shorten turnaround time nowadays packs SOC designs with IP blocks or macros.

4. INTRODUCTION Completely avoiding those routing areas will result in significant underutilization. The over-the-block routing resources with buffering awareness are proposed by [21] and [11].

5. INTRODUCTION Key contributions: – global routing solution with overflows, wirelength, via count and buffering awareness considered simultaneously. – Improve BOB-RSMT (Rectilinear Steiner Minium Tree )algorithm [21] by addressing its two limitations.

6. INTRODUCTION – For any block with overflow, in each iteration we evolve new topologies from inside trees. – Conduct Lagrangian-multipliers-based cost function to reflect the weighted impact from all generated topologies. – An RC-constrained A* search is proposed to help evolve new topologies while meeting slew constraints.

7. PRELIMINARIES A. Basic over-the-block concepts

8. PRELIMINARIES B. Problem Formulation – Wirelength, via cost and total overflow (TOF) are used to evaluate our routing solution. – Over-the-block trees have to satisfy the slew constraints which ensure that every topology has feasible buffering solutions.

9. BOB-ROUTER ALGORITHMS

10. Generate Legal Initial Topologies – an initial RSMT is generated for each net by FLUTE [8]. – BOB-RSMT has two limitations: Movement of the driver for an inside tree is not considered. When two branches at the opposite end of the driver move simultaneously, slew improvement may be underestimated.

11. BOB-ROUTER ALGORITHMS

13. Evolving Legal Congestion-Aware Min-Cost Topologies – 1) Formulations:

14. BOB-ROUTER ALGORITHMS – 2) Pricing the Edges: Calculate price to describe the potential overflow on each edge.

15. BOB-ROUTER ALGORITHMS – 3) Three-level Topology Selection: Level-one: – After we find all inside trees with positive S i, all topologies associated with these unroutable inside trees will be evolved. Level-two: – If evolution of topologies from level-one is unable to keep optimizing the LP formulation, the topologies from level-one, any topology containing overflowed edge(s) will be added.

16. BOB-ROUTER ALGORITHMS Level-three: – If the topology evolution in level-two fails, we evolve topologies covering edges with positive price in addition.

17. BOB-ROUTER ALGORITHMS – 4) RC-constrained A* Search: Evolve new topologies with slew-aware rip-up and reroute. The heuristic cost function we use is the 3-D Manhattan distance.

18. BOB-ROUTER ALGORITHMS C. Outside-tree Routing – After topologies of inside trees are fixed, capacities of all edges within blocks are set to zero before blockage-avoiding outside-tree routing, which will be solved by existing academic routers.

19. EXPERIMENTAL RESULTS BOB-Router – Implemented in the C++ programming language. – On an Intel Core 3.0GHz Linux machine with 16GB memory. – Use 3D global routing benchmarks ISPD 2007 and 2008 Global Routing Contests. – The wire resistance and capacitance for each metal layer are derived from ITRS [1].

20. EXPERIMENTAL RESULTS – use 70ps as our maximal allowed slew.

21. EXPERIMENTAL RESULTS

23. CONCLUSION A new formulation of global routing problem from a different perspective. BOB-Routing can generate slew-violation-free solution with 66% less TOF, 12% less wirelength and 22% less via count compared with the obstacle-avoiding approach.