Field-net QoR Prediction for Circuit Placement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic circuit design processes lack suitable machine learning methods to learn analytical or continuous functions for complex spatial and geometry-related problems like placement, which are essential for predicting quality of results (QoR) during the placement stage, and current methods do not accurately model the geometry of macros and cells within circuit designs.
Innovation Solution
A system and method that utilize principles of field theory and machine learning to determine an analytical function that predicts QoR by sampling field values at various points within a design space, using a neural network to analyze these values and produce a function that describes the field's behavior, allowing for improved circuit design placement and updates during the placement stage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing machine learning methods are used for placement prediction, then computational efficiency is improved, but accuracy in modeling geometry of macros and cells deteriorates
Solution Approach 1:
The patent introduces a field net as an intermediary representation that bridges the gap between discrete circuit elements and continuous spatial modeling. The field net divides the design space into regions influenced by different macros and cells, allowing accurate geometric modeling while maintaining computational efficiency through region-based processing rather than element-by-element analysis.
Solution Approach 2:
The patent segments the design space into multiple regions based on the influence zones of different macros and cells. This segmentation allows the system to model geometric relationships at appropriate scales without processing every individual cell and macro connection, thus achieving both accuracy and efficiency.
2Manufacturing precision
If separate placement optimization is performed, then placement quality is improved, but overall design flow complexity increases
Solution Approach 1:
The patent merges placement optimization with the existing EDA design flow by integrating the field net-based QoR prediction into standard placement tools. This allows placement optimization to be performed as part of the normal design process rather than as a separate complex workflow, reducing overall system complexity while maintaining high placement quality.
Solution Approach 2:
The placement tool uses the field net representation to self-evaluate and self-optimize placement quality without requiring external intervention or complex multi-step optimization processes. The system automatically computes QoR metrics and adjusts placement based on the field net analysis, simplifying the design flow.
3Measurement precision
If detailed geometric modeling is performed, then accuracy of placement prediction is improved, but computing resource consumption increases
Solution Approach 1:
The patent applies local quality analysis by computing field net values only in regions where macros and cells have significant influence. Rather than uniformly processing the entire design space, the system focuses computational resources on areas with complex geometric relationships, achieving accurate predictions with reduced computing resource consumption.
Data Source
AI summary
The present disclosure describes a system and method for determining a function for a quality of results (QoR) for a circuit design. The method includes receiving a circuit design that includes a first node, a second node, and a connection between the first node and the second node in a design space, placing an array of points within the design space, sampling first field values of the circuit design at each point of the array of points, and determining a first sampled function based on the first field values. The method includes moving the array of points within the design space, sampling second field values of the circuit design at each point of the array of points, determining a second sampled function based on the second field values, and determining a function based on the first and second sampled functions. The function produces a QoR for the circuit design.