FAR: Fixed Points Addition & Relaxation Based
PlacementBo Hu and Malgorzata Marek-
Sadowska
University of California,Santa Barbara
Outline Introduction Previous works Fixed points Addition &
Relaxation(FAR) Experimental results Conclusion
Introduction
Placement Methodology
Simulated Annealing Quadratic Partitioning
Partitioning Constant force AR modelFAR
Introduction cont.
Placement Category
Incremental Placement
Detailed Placement Global Placement
FAR
Previous Works Constant Force
“Generic Global Placement and Floorplanning”, Hans Eisenmann and Frank M. Johannes, DAC98.
Attractor-Repeller Model “Attractor-Repeller Approach for Global
Placement”, Hussein Etawil, Shawki Areibi, and Anthony Vannelli, ICCAD-99.
Constant Force(EQ1)
(EQ2)(EQ3)(EQ4)
Vector e is constant force based on density.
Fixed Point A fixed point p(x, y) is a
dimensionless pseudo cell positioned at (x,y) on the chip plane. Fixed point can be anywhere, not
necessarily within the chip boundary. Connectivity between fixed points and real
cells can be dynamically modified.(EQ5)
Category-I: controlling fixed points Controlling fixed points are used to
maintain force equilibrium state.
Controlling Fixed point
Category-II: Perturbing fixed points Perturbing fixed points are used to
transform the placement
Perturbing fixed point
Perturbing fixed point
Perturbing fixed point
Category-III: Constraining fixed points
Constraining fixed points are used to restrict the movement of cells.
constraining fixed point
constraining fixed point
constraining fixed point
Apply FAR in global placement – one transform
initial
CF
FAR
FAR based incremental placement Buffer Site Generation(BSG)
Apply FAR in BSG
Controlling Fixed pointPerturbing fixed pointConstraining fixed point
Experimental results for global placement – wire length
Experimental results for global placement – CPU time for one transformation
Experimental results for BSG – test cases
Experimental results for BSG – CF vs. FAR
Experimental results for BSG
Conclusion FAR based placement approach
can be applied in both global placement( overlapping elimination ) and incremental placement (Buffer Site Generation) contexts.
FAR is quadratic-solver friendly approach.