1 / 39

Robust Window-based Multi-node Technology-Independent Logic Minimization

Robust Window-based Multi-node Technology-Independent Logic Minimization. Jeff L.Cobb W Kanupriya Gulati D Sunil P. Khatri D. W Texas Instruments, Inc. D Dept. of ECE, Texas A&M University. Overview Introduction Background Previous work Approach Experimental results Conclusions.

viviana
Télécharger la présentation

Robust Window-based Multi-node Technology-Independent Logic Minimization

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. Robust Window-based Multi-node Technology-Independent Logic Minimization Jeff L.Cobb W Kanupriya Gulati D Sunil P. Khatri D W Texas Instruments, Inc. D Dept. of ECE, Texas A&M University

  2. Overview Introduction Background Previous work Approach Experimental results Conclusions

  3. Introduction VLSI design flow • HDL (Verilog, VHDL) • Logic optimization • Physical design

  4. Introduction • Purpose of logic optimization • Reduce area • Reduce power • Reduce delay • Logic optimization • Technology-independent optimization • Goal: reduce literal count • Technology-dependent optimization

  5. Background • Don’t Cares • Logic function allowed to have 0 or 1 as possible output for a given input ODC SDC • XDC: External don’t cares given

  6. Background • Don’t Cares • Computed for one node at a time • Cannot capture multi-node flexibility xy (x+y) = xy(xy) + (x+y)(x+y) = xy+xy = x y • Goal: multi-node logic minimization • Yields a Boolean relation • Need to determinize this relation for solution

  7. Background • Boolean relations • Can express more than one allowed output vector for a single input vector • Don’t cares only express flexibility for a single output

  8. Terminology

  9. Problem Definition • Implement dual-node Boolean relation-based multi-level logic minimization technique • Goals: • Method must scale to large designs • Compare to best don’t care-based method (single-node)

  10. Previous Work • [CM77] Formulated multi-node minimization problem • No results provided • [WW94] Multi-node minimization • Extremely large runtimes, works on very small designs • [MB05] Single node approach, uses windowing and SAT based formulation • Used for comparison purposes • This work: Efficient choice of nodes, window based, efficient quantification scheduling 10

  11. Approach • Key features • Dual node optimization • Careful node pair selection • Window based optimization technique • Early quantification for efficiency

  12. Approach

  13. Node Pair Selection • Node Pair Selection

  14. Node Pair Selection

  15. Node Pair Selection

  16. Node Pair Selection • Compute common input ratio • Compute common output ratio • Select node pairs that satisfy

  17. Subnetwork Extraction

  18. Subnetwork Extraction

  19. Building the Relation • where

  20. Quantification Scheduling

  21. Quantification Scheduling

  22. Quantification Scheduling

  23. Endgame • Call BREL (a Boolean relation minimizer) to minimize • Returns new nodes and • Graft new nodes into • Delete original nodes • ,

  24. BREL • BREL is a heuristic Boolean relation solver • Solving a Boolean relation • Same as minimum cost determinization of the relation (i.e. finding the lowest cost function which is contained in the relation) • Branch and bound approach

  25. Experimental Results • Implemented in SIS • Uses CUDD ROBDD Package • 15 benchmark circuits from mcnc91, itc99 • Metric for quality: literal count • Preprocessing steps: • Removes constant-valued nodes • Removes nodes that do not fanout • Merges functionally identical nodes

  26. Experimental Results • Parameter selection • 4 parameters to node selection algorithm • Goal: Find “golden” values

  27. Experimental Results • Parameter:

  28. Experimental Results • Parameters: • : Window size • : Partners for

  29. Experimental Results

  30. Experimental Results • Parameter:

  31. Experimental Results • “Golden” parameter values: • Can be modified to balance quality/runtime

  32. Experimental Results • Compared versus • 12% lit. improvement • 38x runtime increase • But runtimes are • still within 3-4 min • Low memory (#BDD • nodes) • High gain (number of • node pairs which give • an improvement)

  33. Experimental Results • Run after • 13% lit. improvement • Both use 2x2 windows

  34. Experimental Results • Limit subnetwork size τ

  35. Conclusions • 12% less literals than best DC approach • Runtimes under 4 minutes for largest network • Low memory usage • Further reduce literals by 13% after running best DC approach • Future Work • Consider 3+ nodes in relation • SAT-based relation construction • Alternative to BREL

  36. Thank you!

  37. SAT-Sweep

  38. BREL

  39. BREL

More Related