Learning-Based Flexible Block Shaping for IC Placement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In integrated circuit (IC) design, existing methods struggle to efficiently shape flexible blocks with flexible aspect ratios on a chip canvas, often resulting in area violations and congestion due to overlapping bounding boxes, which complicates power, performance, and area (PPA) optimization.
Innovation Solution
A learning-based approach using a neural network and convex optimization is employed to reshape flexible blocks, minimizing overlapping areas and adhering to aspect ratios, while maintaining relative positions and optimizing wire length, through iterative updates of bounding box areas and shapes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If analytical methods or reinforcement learning are used for BBox placement, then positioning can be achieved, but overlapping occurs resulting in area violations and congestion
Solution Approach 1:
The patent performs preliminary actions by first identifying overlapping regions between BBoxes and blockages, then proactively resolving them through iterative convex optimization before final placement is completed. This prevents area violations from occurring in the first place rather than correcting them afterward.
Solution Approach 2:
The patent employs dynamic iterative convex optimization that continuously adjusts BBox dimensions and positions based on real-time feedback about overlaps and area constraints. The aspect ratios are dynamically updated across iterations to converge on an optimal non-overlapping configuration that satisfies all constraints.
2Object-generated harmful factors
If BBox areas are expanded to avoid overlap, then congestion is reduced, but area requirements are violated
Solution Approach 1:
The patent changes parameters by iteratively adjusting BBox aspect ratios and dimensions through convex optimization. Instead of simply expanding areas, the system modifies multiple parameters simultaneously (width, height, position) while maintaining aspect ratio constraints, thereby reducing overlaps without violating area requirements.
Solution Approach 2:
The patent implements feedback mechanisms where each iteration evaluates the current placement against area constraints and overlap conditions. The convex optimization uses this feedback to compute updated aspect ratios and positions that progressively reduce congestion while ensuring area compliance in the final solution.
3Productivity
If block shapes are adjusted to be rectangle-like without zig-zag edges, then efficiency is improved, but the shaping process becomes exceptionally difficult
Solution Approach 1:
The patent segments the complex block shaping problem into manageable convex optimization subproblems. By dividing the shaping process into iterative steps that each handle specific constraints (area, aspect ratio, non-overlap), the system achieves rectangle-like shapes efficiently without being overwhelmed by the overall complexity.
Solution Approach 2:
The patent replaces traditional mechanical or geometric shaping methods with convex optimization mathematics. This substitution transforms an exceptionally difficult geometric problem into a computationally tractable optimization problem that can be solved systematically using standard convex optimization algorithms.
4Manufacturing precision
If iterative updates are performed to minimize overlapping, then area compliance is achieved, but computation time increases
Solution Approach 1:
The patent applies partial action by performing a limited number of iterative convex optimization steps rather than exhaustive search. Each iteration makes partial progress toward the optimal solution, and the process terminates when convergence criteria are met, achieving sufficient area compliance without unnecessary computation time.
Solution Approach 2:
The patent uses beforehand cushioning by initializing BBox aspect ratios and positions strategically before the iterative process begins. This preliminary setup reduces the distance to the optimal solution, allowing the convex optimization to converge faster and reducing overall computation time while still achieving area compliance.
Data Source
AI summary
The application discloses a method and a system for shaping flexible blocks on a chip canvas in an integrated circuit design. An input is received describing geometric features of flexible blocks. A set of flexible blocks are generated based on the input. Obtained block areas of the set of flexible blocks are computed. Whether the set of flexible blocks are legal is determined based on determining whether area differences between the obtained block areas and a plurality of required areas for the set of flexible blocks meet a requirement. The set of flexible blocks are updated until the set of flexible blocks are all legal.