EDA Parameter Search Filtering for Black-Box Circuit Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic design automation (EDA) systems face challenges in optimizing black-box functions due to complex, nonlinear, and noisy relationships between input parameters and outputs, leading to inefficient parameter space exploration, computational expense, and suboptimal results, particularly in circuit design optimization.
Innovation Solution
Implementing an interposer system with surrogate models to filter and evaluate parameter values before passing them to the black-box function, using feedback to guide machine learning (ML) models and reduce unnecessary computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If the ML model explores the full parameter space to find optimal values, then the quality of circuit design optimization improves, but the computational expense and time required increase significantly
Solution Approach 1:
The interposer system performs preliminary evaluation of parameter combinations using surrogate models before submitting them to the expensive black-box EDA tool. This preliminary filtering action identifies and rejects unpromising parameter combinations early, preventing waste of computational time on futile explorations while maintaining thorough search of promising regions.
Solution Approach 2:
The interposer system acts as an intermediary layer between the ML model and the black-box EDA tool. It uses surrogate models to approximate the black-box function's behavior, providing a computationally efficient filter that guides the ML model's parameter generation without requiring direct, expensive evaluations of the actual EDA tool for every parameter combination.
2Measurement precision
If the system evaluates all generated parameter values through the black-box function, then the accuracy of result evaluation improves, but the computational expense increases significantly
Solution Approach 1:
The system performs partial evaluation by using surrogate models to assess parameter combinations before full black-box evaluation. This partial filtering approach evaluates only the most promising parameter combinations with the expensive black-box function, achieving sufficient evaluation accuracy for practical optimization while dramatically reducing computational energy consumption by avoiding exhaustive evaluation of all generated parameters.
3Adaptability or versatility
If the ML model generates parameter values without guidance, then the diversity of explored solutions improves, but the efficiency of finding desirable parameter combinations decreases
Solution Approach 1:
The interposer system implements feedback by evaluating parameter combinations with surrogate models and providing guidance signals to the ML model. When parameter combinations violating design rules or failing to meet objectives are detected, the system feeds this information back to the ML model, which adjusts its generation strategy to produce more desirable parameter combinations while maintaining solution diversity through controlled exploration.
Data Source
AI summary
A set of parameter values may be generated by a machine learning (ML) model, where the set of parameter values may be used by a black-box function to generate a set of outputs based on a set of inputs. It may be determined whether the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate one or more desired goals. If so, a first response may be provided to the ML model that discourages the ML model from generating sets of parameter values that are similar to the set of parameter values. Otherwise, the set of parameter values may be provided to the black-box function, a second response may be determined based on the set of outputs generated by the black-box function, and the second response may be provided to the ML model.