Floorplanning with I/O Assignment via Feasibility-Seeking and Superiorization Methods.

Area of Science:

• Computer Science
• Electrical Engineering
• Algorithm Design

Background:

• Floorplanning problems involve complex, heterogeneous constraints that challenge conventional algorithms.
• Existing projection-based feasibility-seeking methods lack guaranteed convergence and are sensitive to initialization.
• Algorithmic divergence and oscillation are significant issues in current feasibility-seeking approaches.

Purpose of the Study:

• To develop a robust and efficient feasibility-seeking framework for complex floorplanning problems.
• To improve the convergence properties and initialization sensitivity of projection-based algorithms.
• To integrate superiorization methods for bridging feasibility-seeking and constrained optimization in floorplanning.

Main Methods:

• A quantitative property for optimal initial point selection was identified and analyzed.
• A resetting strategy was implemented to mitigate algorithmic divergence in projection-based methods.
• The superiorization method (SM) was applied to steer iterations towards feasible solutions with reduced wirelength.

Main Results:

• The proposed Per-RMAP achieved legal floorplanning 166x faster than branch-and-bound (B&B) with a 5% wirelength increase.
• Incorporating I/O assignment constraints yielded a 6% wirelength improvement.
• For soft modules, a 15% runtime improvement was observed compared to the state-of-the-art analytical method PeF.

Conclusions:

• The novel algorithmic flow effectively handles diverse floorplanning constraints, including I/O assignment and soft modules.
• The approach demonstrates a significant improvement in runtime efficiency while maintaining competitive solution quality.
• This work provides a balanced and adaptable solution for complex, real-world floorplanning challenges.