Machine-Trained Network for IC Physical Design Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Integrated circuit (IC) design layouts often fail to account for manufacturing constraints, leading to unreliable production and the need for post-design modifications when sent to fabrication foundries.
Innovation Solution
The use of machine-trained networks to predict how a physical IC design will appear in manufacturing, allowing for real-time modifications to meet design considerations and constraints, thereby improving the quality and efficiency of the design process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If IC design layouts are created without considering manufacturing constraints, then the design process is faster and simpler, but the manufacturing reliability deteriorates and post-design modifications are required
Solution Approach 1:
The patent applies preliminary action by integrating manufacturing constraint checks and machine learning-based manufacturing outcome predictions into the physical design stage itself. This allows designers to identify and correct manufacturing issues before sending layouts to fabrication, eliminating the need for post-design modifications while maintaining design process efficiency.
Solution Approach 2:
The patent implements feedback mechanisms where machine learning models predict manufacturing outcomes based on physical design inputs, and this predicted information feeds back into the design process to guide modifications. This closed-loop feedback ensures manufacturing constraints are met while maintaining design productivity.
2Device complexity
If traditional physical design tools are used without machine learning integration, then the design process is simpler, but the number of iterations between design and manufacturing stages increases
Solution Approach 1:
The patent introduces machine learning models as intermediary components between physical design tools and manufacturing processes. These ML models predict manufacturing outcomes and guide design modifications, reducing the number of iterations required without significantly increasing overall system complexity.
Solution Approach 2:
By performing manufacturing outcome predictions and constraint checks in advance during the physical design stage, the system eliminates the need for multiple back-and-forth iterations between design and manufacturing stages, significantly reducing iteration time.
3Reliability
If machine learning models are integrated into physical design tools, then manufacturing reliability improves, but the device complexity increases
Solution Approach 1:
Machine learning models serve as intermediary components that bridge physical design tools and manufacturing processes. These ML models handle the complexity of manufacturing predictions internally, allowing the overall system to achieve improved manufacturing reliability without requiring fundamental changes to the core design tool architecture.
Data Source
AI summary
A method of some embodiments receives an initial first physical design of a circuit. The method uses a machine-trained network to generate a second physical design that is a prediction of how the first physical design will look at a subsequent manufacturing stage. The method then uses the second physical design to modify the first physical design. Examples of such modifications include modifying a set of one or more routes in the first physical design and/or modifying a set of placement locations of a set of one or more sub-circuits or circuit components defined in the first physical design.