VLSI circuit placement with rectilinear modules using three-layer force-directed self-organizing maps.

Area of Science:

• Computer Science
• Artificial Intelligence
• VLSI Design

Background:

• Circuit placement is a complex optimization problem, especially with arbitrarily shaped rectilinear modules.
• Existing methods like simulated annealing can be time-consuming and may not yield optimal results.

Purpose of the Study:

• To introduce a novel three-layer force-directed self-organizing map for rectilinear module placement.
• To simultaneously minimize wire length and module overlap in circuit design.

Main Methods:

• A three-layer neural network architecture with an additional hidden layer is proposed.
• Hidden neurons model rectilinear modules by representing partitioned rectangles.
• Collective computation among hidden neurons facilitates module interaction and placement convergence.

Main Results:

• The proposed model effectively resolves the rectilinear module placement problem.
• Placement results demonstrate a reduction in both wire length and module overlap.
• The method shows superior performance compared to simulated annealing in terms of total wire length.

Conclusions:

• The developed neural network model offers an efficient and effective solution for rectilinear module placement.
• The approach successfully balances contradictory optimization criteria.
• Further investigation into parameter tuning can yield even better placement solutions.