Machine Learning Floorplan Placement Tool for Incremental NoC Topology Editing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Creating an optimal network-on-chip (NoC) topology on a floorplan that meets performance requirements such as connectivity, latency, and power consumption is a complex and time-consuming task, often requiring frequent redesigns due to changes in chip floorplan or IP components, and existing tools lack efficient interactive computation and machine learning for legal and optimized placement.
Innovation Solution
A design tool that utilizes machine learning models for interactive computation of legal and optimized network-on-chip placement on a floorplan, employing a two-phase process of global coarse-grained and local fine-grained placement, with real-time feedback and training to ensure compliance with clock and power domain constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional manual or automated placement tools are used to create optimal NoC topology on floorplan, then placement quality and performance requirements are met, but the design process becomes extremely time-consuming and requires frequent complete redesigns
Solution Approach 1:
The system performs preliminary placement actions by predicting the optimal position of network elements before the user finalizes the topology design. The machine learning model pre-computes placement suggestions based on the current floorplan state, allowing users to see predicted placements immediately when adding or removing elements, thus avoiding time-consuming iterative redesigns
Solution Approach 2:
The system implements real-time feedback by continuously monitoring changes in the NoC topology (addition/removal of elements, floorplan modifications) and immediately re-computing predicted placements using the machine learning model. This feedback loop provides users with up-to-date placement suggestions that adapt to any topology changes, eliminating the need for manual re-optimization
2Adaptability or versatility
If the NoC topology is edited interactively by adding or removing elements, then adaptability and user control are improved, but maintaining legal and optimized placement becomes computationally complex and slow
Solution Approach 1:
The machine learning model performs self-service by automatically detecting topology changes and re-computing optimal placements without requiring user intervention or complex computational workflows. The system monitors itself for changes in the NoC topology and autonomously updates placement predictions, simplifying the interaction while maintaining optimization quality
Solution Approach 2:
The system efficiently handles topology edits by detecting changes in key parameters (number of network elements, floorplan geometry, connectivity requirements) and re-training or fine-tuning the machine learning model only for the affected regions or parameters. This selective parameter update approach maintains adaptability while reducing computational complexity compared to complete re-optimization
3Productivity
If existing placement tools are used without machine learning, then computational simplicity is maintained, but the ability to provide fast interactive computation and optimized placement for incremental updates is insufficient
Solution Approach 1:
The machine learning model is segmented into modular components that can independently process different aspects of placement prediction. The system divides the floorplan into regions and processes placement predictions for each region separately, allowing parallel computation and faster overall performance while managing the complexity of the machine learning integration
Solution Approach 2:
The system uses pre-trained machine learning models that capture placement patterns and optimizations from extensive training data. Instead of computing optimal placements from scratch each time, the system copies learned placement strategies and adaptations from the trained model to new topology configurations, significantly accelerating computation while handling the complexity through model reusability
Data Source
AI summary
A design tool is disclosed having a machine learning model for automatic physical implementation guidance that allows interactive computation of a legalization and optimization placement of a network-on-chip (NoC) topology on a floorplan. The machine learning model performs one or more iterations during NoC topology editing. The machine learning model also includes the ability to use feedback to train the machine learning model for automated and assisted topology analysis and synthesis. The tool generates physical implementation guidance, which is during physical implementation of the synthesized NoC.