ML-Based Analog Device Placement for EDA
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic circuit design systems face challenges in automatically placing analog devices at graphical user interfaces to meet snapping, row style, and grouping requirements, leading to inefficient optimization processes due to overlapping placements and difficulties in predicting device grouping and isolation.
Innovation Solution
A computer-implemented method and system that utilize machine learning models to predict device placements and create constraints based on electronic design schematics and layouts, including predictions for row placement, position, and grouping within rectangular shapes, using cost functions related to circuit area and wirelength.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional optimization methods are used for device placement, then the placement can be generated, but the optimization process is inefficient due to overlapping placements and extensive manual effort required
Solution Approach 1:
The system performs preliminary actions by training machine learning models on existing layout data before the actual placement optimization. The models learn grouping patterns, row style requirements, and snapping constraints in advance, enabling them to predict optimal placements without extensive manual optimization during the placement process itself
Solution Approach 2:
Machine learning models serve as intermediaries between the design requirements and the placement optimization process. The models translate learned patterns from training data into predictions that guide the placement algorithm, mediating between raw design constraints and optimized placement results
2Manufacturing precision
If all possible device placement instances are formulated as optimization variables, then complete placement coverage is achieved, but the design space becomes infeasible due to overlapping placements
Solution Approach 1:
The system performs preliminary grouping predictions using trained machine learning models before formulating the optimization problem. By predicting which devices should be grouped together in advance, the system reduces the number of optimization variables needed, making the design space feasible while maintaining placement accuracy
Solution Approach 2:
The placement problem is segmented into multiple independent sub-problems: grouping prediction, row style prediction, and final placement optimization. Each segment handles a specific aspect of the placement task, reducing overall complexity while maintaining comprehensive coverage
3Reliability
If manual methods are used to predict device grouping and isolation requirements, then accurate placement constraints are obtained, but extensive manual effort is required
Solution Approach 1:
The system implements self-service by automatically training machine learning models on existing layout data to learn grouping patterns and isolation requirements. The models then autonomously predict these requirements for new designs without requiring manual specification, maintaining reliability while eliminating extensive manual effort
Solution Approach 2:
The system uses feedback from training data consisting of existing layouts with known grouping and isolation patterns. The models learn from this feedback to improve their predictions, continuously refining their ability to accurately predict device grouping and isolation requirements
Data Source
AI summary
The present disclosure relates to a computer-implemented method for electronic design. Embodiments may include receiving, using at least one processor, an electronic design schematic and an electronic design layout and training a model using at least one predictor associated with the electronic design layout. Embodiments may further include obtaining an updated model, based upon, at least in part, the training. Embodiments may also include applying the updated model to a second electronic design schematic or a second electronic design layout, wherein one or more hard constraints or one or more soft constraints or both are created, based upon, at least in part, the model.