Placement-Centered Research Directions and New Problems - PowerPoint PPT Presentation

About This Presentation
Title:

Placement-Centered Research Directions and New Problems

Description:

Could focus on critical paths that can be easily fixed by logic changes ... There are always tradeoffs between placement objectives ... – PowerPoint PPT presentation

Number of Views:19
Avg rating:3.0/5.0
Slides: 12
Provided by: celine4
Learn more at: https://vlsicad.ucsd.edu
Category:

less

Transcript and Presenter's Notes

Title: Placement-Centered Research Directions and New Problems


1
Placement-Centered ResearchDirections and New
Problems
  • Xiaojian Yang
  • Amir Farrahi
  • Synplicity Inc.

2
Topics
  • Basic Requirements
  • Mixed Block Standard-Cell Placement
  • Incremental Placement
  • Placement-In-Flow
  • Congestion/Routability
  • Timing Driven Placement
  • Other objectives

3
Basic Requirements
  • Large scale problem
  • Fixed-die with 50-80 density
  • Wirelength
  • Basic algorithms and methodologies
  • Min-cut, Analytical, Annealing
  • Clustering
  • Multi-level
  • Hierarchical

4
Placement with white space
MCNC benchmarks
Industrial designs
5
Mixed Mode Placement
  • Fixed Macros
  • Non-rectangle placeable area
  • Min-cut and annealing can be adjusted. What about
    quadratic approach?
  • Movable Macros
  • Floorplanning macros then place standard-cells
  • Clustering and floorplanning hard/soft macros
  • Splitting hard macros to form standard-cell
    placement
  • Modified partitioning --- cut hard macros
  • Low utilization --- not packing-type
    floorplanning

6
Incremental Placement
  • Caused by various reasons
  • Placement based logic optimization
  • Clock tree construction
  • Power rail / special nets construction
  • ECO
  • Incremental Detailed Placement
  • Highly constrained
  • Minimum disturbance if the change is small
  • Appropriate replacement based on the amount of
    change
  • Measurement of the disturbance

7
Placement-In-Flow
  • Floor-Placer (Steve Teig, ISPD2002)
  • Mixed macro standard-cell placement
  • Floorplan with unfixed soft macros
  • Placement with unfixed hard macros
  • Placement with Global Routing
  • Change placement based on global routing results
  • Placement in Physical Synthesis
  • Choose the right optimization in placement
  • Priorities of placement objectives
  • Positions of Global/Detailed placement
  • Interleaving logic synthesis optimizations

8
Congestion/Routability
  • Transparent congestion management
  • Caldwell/Kahng/Markov DAC2000
  • Congestion improvement if and only if required
  • Tradeoff between congestion and wirelength/delay
  • Use of White space / Free space
  • Uniform distribution
  • routability-driven distribution (not too
    aggressive)
  • Routing resource aware placement
  • Horizontal/vertical wires
  • Blockages in routing layers

9
Timing
  • Avoid pure timing-driven placement
  • May hurt other objectives
  • Could focus on critical paths that can be easily
    fixed by logic changes
  • Focus timing-driven placement with minor changes
    on the netlist
  • Buffer insertion, Gate sizing, Gate replication
  • Transparent timing optimization
  • Again, tradeoff between delay and routability
  • Constraint-driven rather than minimization
  • Understanding the relationship between
    routability and timing
  • Negative reducing congestion increases critical
    path
  • Netural reducing congestion does not affect
    critical path
  • Positive reducing congestion brings better
    post-routing delay

10
Other Objectives
  • Power Driven Placement
  • Multi-voltage designs
  • Signal Integrity Cross talk driven placement
  • IR-Drop Issue and Placement
  • Thermal Issue and Placement
  • Placement Constraints
  • High Density Design Placement
  • Multiple objectives in placement

11
Summary
  • Placement remains a challenging problem, and
    becomes more difficult with more constraints
  • There are always tradeoffs between placement
    objectives
  • Placement combined with other optimizations could
    gain more
  • Benchmarks are indispensable for academic research
Write a Comment
User Comments (0)
About PowerShow.com